Handbook of Neuroprosthetic Methods - W. Finn, P. LoPresti (CRC, 2003) WW

1100_FM.fm Page iv Friday, January 31, 2003 1:54 PM

This book contains information obtained from authentic and highly regarded sources. Reprinted materialis quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonableefforts have been made to publish reliable data and information, but the author and the publisher cannotassume responsibility for the validity of all materials or for the consequences of their use.

Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronicor mechanical, including photocopying, microfilming, and recording, or by any information storage orretrieval system, without prior permission in writing from the publisher.

All rights reserved. Authorization to photocopy items for internal or personal use, or the personal orinternal use of specific clients, may be granted by CRC Press LLC, provided that $1.50 per pagephotocopied is paid directly to Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923USA. The fee code for users of the Transactional Reporting Service is ISBN 0-8493-1100-4/03/$0.00+$1.50. The fee is subject to change without notice. For organizations that have been granteda photocopy license by the CCC, a separate system of payment has been arranged.

The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, forcreating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLCfor such copying.

Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and areused only for identification and explanation, without intent to infringe.

Visit the CRC Press Web site at www.crcpress.com

© 2003 by CRC Press LLC

No claim to original U.S. Government worksInternational Standard Book Number 0-8493-1100-4

Library of Congress Card Number 2002031437Printed in the United States of America 1 2 3 4 5 6 7 8 9 0

Printed on acid-free paper

Library of Congress Cataloging-in-Publication Data

Handbook of neuroprosthetic methods / edited by Warren E. Finn, Peter G. LoPresti.p. ; cm. -- (Biomedical engineering series)

Includes bibliographical references and index.ISBN 0-8493-1100-4 (alk. paper)1. Neural stimulation--Handbooks, manuals, etc. 2. Prosthesis--Handbooks, manuals,

etc. 3. Cochlear implants--Handbooks, manuals, etc. 4. Neurons--Handbooks, manuals,etc. I. Finn, Warren E. II. LoPresti, Peter G. III. Biomedical engineering series (BocaRaton, Fla.)

[DNLM: 1. Nervous System Diseases--rehabilitation. 2. Prosthesis and Implants. 3.Electric Stimulation Therapy--methods. 4. Nervous System Physiology. 5. ProsthesisDesign. WL 140 H2362 2002]RC350.N48 H36 2002616.8′046--dc21 2002031437

©2003 CRC Press LLC

http://www.crcpress.com/

1100_FM.fm Page v Friday, January 31, 2003 1:54 PM

Preface

PurposeThe purpose of The Handbook of Neuroprosthetic Methods is threefold. First,the book combines the most commonly employed concepts, applications,and knowledge from the many disciplines associated with neuroprostheticresearch in a clear and instructive way. Second, the book provides examplesof neuroprosthetic systems at different stages of development, from the moremature cochlear implant to the maturing areas of upper-limb and motorcontrol to the relatively fledgling area of visual prostheses. The book exploresthe varying developmental processes to give the reader guidance on issuesthat have yet to be solved, successful strategies for solving such problems,and the potential pitfalls encountered when developing neural prostheses.Third, the book introduces key topics at a level that is useful to both newand practicing professionals working directly or indirectly with neuropros-thesis projects. In this way, the book provides an accessible common groundand perhaps fosters a more effective and productive collaborative environ-ment for multidisciplinary teams working on protheses.

OrganizationThe book is organized into six main sections, starting with basic neurophys-iology and ending with some of the emerging technologies that will havesignificant impact on the next generation of prostheses. Section I providesan overview of the significant events in the field of neuroprostheses and thebroad problems that remain a challenge to the development of a functionaland practical neuroprosthetic system. Section II addresses the main targetof the neuroprostheses, the neuron. The main topics in this section are howneurons can become electrically excited to produce signals used to restoresensory or motor function and the ways this behavior can be modeledmathematically and predicted.

Section III addresses the important and difficult task of recording fromand stimulating neurons, either in the laboratory or in devices implanted inhumans. The book first addresses the design problem of finding an effectivestimulus to adequately activate a population of neurons in a nerve pathwayto restore function. Of importance here is the concept of not only sending


the correct signals but also sending the signals in a way that does not injure

1100_FM.fm Page vi Friday, January 31, 2003 1:54 PM

or destroy the nerve cells with which one is communicating. Second, thebook addresses the components and devices commonly used for communi-cation in current systems. Third, the book addresses the task of listening tothe chatter amongst neurons using many of the same devices utilized forsignal transmission.

In Section IV, the book addresses some issues related to processing therecorded neural signals. The goal of signal processing is to understand howthe neurons are communicating and how information is encoded within theircommunications. Section V provides examples of three neuroprosthetic sys-tems at different stages of the development cycle. Section VI introduces someemerging technologies that promise to alter current approaches to neuro-prosthetic design. Through this organization, the book accentuates the poten-tial contribution of the biological and engineering fields brought together tosolve the complex problems that are at the heart of the neuroprosthetic field.

Each chapter follows a basic format that we hope the reader findsuseful. Each chapter begins with a brief history of the topic and thenaddresses the fundamental issues and concepts. It is hoped that the mannerin which each topic is addressed provides understanding to the beginningpractitioner as well as guidance to practitioners with more experience inthe field. The last part of each chapter provides practical applications andexamples that relate the topic to the actual design and implementation ofa neuroprosthetic system or device. In this way, each chapter provides aconnection between theory and practice that will help the reader bettercomprehend the material presented.

Warren FinnTulsa, Oklahoma

Peter LoPrestiTulsa, Oklahoma


1100_FM.fm Page vii Friday, January 31, 2003 1:54 PM

The editors

Warren E. Finn, Ph.D., is an Associate Professorof Medical Physiology in the Department ofPharmacology and Physiology in the Biomedi-cal Sciences Program at the Oklahoma StateUniversity Center for Health Sciences. Dr. Finnearned his Baccalaureate and Masters of Sci-ence degrees in zoology at University of Wis-consin. In addition, he earned his Ph.D. in thebiological sciences, physiology from TexasA&M University in College Station, TX. Dr.Finn teaches medical and graduate students inthe fields of cellular, molecular and integrativeneurophysiology. His research interests are inthe areas of cell culturing and the electrophys-iology of retinal neurons. He co-coordinateswith Dr. LoPresti the Artificial Vision Project, a

multidisciplinary research program studying vision neuroprosthetics. Thisproject provides research training for students in medicine, biomedical sci-ences, and electrical engineering. Dr. Finn has worked on various bioengi-neering projects, such as the biophysics of hypothermia as a treatment forcerebral ischemia, the activation of sensory neurons in myocardial ischemia,and the electrophysiology of amblyopic eye disease. With Dr. LoPresti, hehas authored studies on animal models for the development of retinal pros-theses reported at various IEEE Engineering in Medicine and Biology SocietyAnnual International Conferences. Dr. Finn contributed three chapters to theHandbook of Endocrinology, published by CRC Press. He is an active par-ticipant in policy and strategic planning in biomedical engineering throughhis activities in the development of intellectual property policies and tech-nology transfer.


Peter G. LoPresti, Ph.D.,

an associate professor

1100_FM.fm Page viii Friday, January 31, 2003 1:54 PM

of electrical engineering at the University ofTulsa, teaches graduate and undergraduatecourses in electronics, signal processing, andoptical communications. He earned a Ph.D. inelectrical engineering from the PennsylvaniaState University and a B.S. in electrical engi-neering from the University of Delaware. Hiscurrent research interests include visual neuro-prosthetics, fiber-optic sensors, and optical net-working, and he serves as the director for theWilliams Communications Fiber-Optic Net-working Laboratory at the University of Tulsa.Dr. LoPresti is highly dedicated to the improv-ing the quality of engineering education, hav-ing won both university and departmental

teaching awards and serving as coordinator for the electrical engineeringcomponent of the Tulsa Undergraduate Research Challenge program, whichprovides accelerated learning and research experiences for undergraduates.


1100_FM.fm Page ix Friday, January 31, 2003 1:54 PM

Contributors

David J. AndersonUniversity of Michigan Ann Arbor, Michigan

Danny BanksMonisys, Ltd.Birmingham, England

Steven BarnesDalhousie UniversityHalifax, Nova Scotia, Canada

Rizwan BashirullahNorth Carolina State UniversityRaleigh, North Carolina

Chris DeMarcoNorth Carolina State UniversityRaleigh, North Carolina

Kenneth J. Dormer The University of Oklahoma Health

Sciences CenterOklahoma City, Oklahoma

Kevin EnglehartInstitute of Biomedical Engineering,

and Department of Electrical and Computer Engineering

University of New BrunswickFredericton, New Brunswick,

Canada

Warren E. FinnOklahoma State UniversityTulsa, Oklahoma

Robert J. GreenbergSecond Sight, LLCValencia, California

Warren M. GrillCase Western Reserve UniversityCleveland, Ohio

Jamille F. HetkeUniversity of MichiganAnn Arbor, Michigan

Bernard HudginsInstitute of Biomedical Engineering,



Canada

Mark S. HumayunDoheny Retina InstituteUniversity of Southern CaliforniaLos Angeles, California

Richard T. LauerShriners Hospitals for ChildrenPhiladelphia, Pennsylvania


Dongchul C. Lee

P. Hunter Peckham

1100_FM.fm Page x Friday, January 31, 2003 1:54 PM

Case Western Reserve UniversityCleveland, Ohio

Wentai LiuNorth Carolina State UniversityRaleigh, North Carolina

Peter G. LoPresti University of TulsaTulsa, Oklahoma

Cameron C. McIntryeThe Johns Hopkins UniversityBaltimore, Maryland

Richard A. NormannUniversity of UtahSalt Lake City, Utah

Philip ParkerInstitute of Biomedical Engineering,



Canada

C. PearsonUniversity of DurhamDurham, England

Rehabilitation Engineering CenterMetroHealth Medical CenterCleveland, Ohio

Michael C. PettyUniversity of DurhamDurham, England

Frank RattayTU-BioMedVienna University of TechnologyVienna, Austria

Susanne ResatzTU-BioMedVienna University of TechnologyVienna, Austria

Donald L. Russell Carleton UniversityOttawa, Ontario, Canada

Praveen SinghNorth Carolina State UniversityRaleigh, North Carolina

David J. WarrenUniversity of UtahSalt Lake City, Utah

James D. WeilandDoheny Retina InstituteUniversity of Southern CaliforniaLos Angeles, California


1100_FM.fm Page xi Friday, January 31, 2003 1:54 PM

Acknowledgments

Dr. Finn would like to thank his wife, Judith, and daughters, Kirstin andArikka, for their help, support, and encouragement during this writingproject. They are all excellent writers in their own right, and he greatly valuedtheir advice during these months. He would also like to thank his colleaguesDavid John and George Brenner for their encouragement and support inundertaking this project and the sharing of their wisdom as authors.

Dr. LoPresti would like to acknowledge the unwavering support of hisfamily, both natural and adopted, through the process of developing thisbook. In particular, he acknowledges the support of Carrie and Joshua, whokeep his spirits up and his priorities straight.

Drs. Finn and LoPresti both wish to thank their many students in med-icine and electrical engineering who have participated in the Artificial VisionProject over the years. They appreciate their many hours of helpful discus-sion and persistence in the laboratory. They would also like to thank DavidMooney, Assistant Librarian at the Oklahoma State University Center forHealth Sciences, for the generous sharing of his knowledge and search skillsof the world’s literature. They also thank Jeffrey Shipman for his assistancewith the historical timeline. They also wish to acknowledge the excellenteditorial assistance of Susan Farmer, Helena Redshaw, Robert Stern, SusanFox, and many others at CRC Press in making this book a reality.


1100_BookFMTOC.fm Page xiii Friday, January 31, 2003 2:09 PM


Table of Contents

Section I: Introduction

Chapter 1Introduction to neuroprostheticsWarren E. Finn and Peter G. LoPresti

Section II: Neurons and Neuron Modeling

Chapter 2Neuronal excitability: membrane ion channelsSteven Barnes

Chapter 3Neuron modelingFrank Rattay, Robert J. Greenberg, and Susanne Resatz

Section III: Stimulating and Recording of Nerves and Neurons

Chapter 4Stimulating neural activityJames D. Weiland, Mark S. Humayun, Wentai Liu, and Robert J. Greenberg

Chapter 5Extracellular electrical stimulation of central neurons: quantitative studiesDongchul C. Lee, Cameron C. McIntyre, and Warren M. Grill

Chapter 6Semiconductor-based implantable microsystemsWentai Liu, Praveen Singh, Chris DeMarco, Rizwan Bashirullah, Mark S. Humayun, and James D. Weiland

Chapter 7

1100_BookFMTOC.fm Page xiv Friday, January 31, 2003 2:09 PM


Silicon microelectrodes for extracellular recordingJamille F. Hetke and David J. Anderson

Section IV: Processing Neural Signals

Chapter 8Wavelet methods in biomedical signal processingKevin Englehart, Philip Parker, and Bernard Hudgins

Chapter 9Neuroprosthetic device designDonald L. Russell

Section V: Prosthetic Systems

Chapter 10Implantable electronic otologic devices for hearing rehabilitationKenneth J. Dormer

Chapter 11Visual neuroprosthesesDavid J. Warren and Richard A. Normann

Chapter 12Motor prosthesesRichard T. Lauer and P. Hunter Peckham

Section VI: Emerging Technologies

Chapter 13Neurotechnology : microelectronicsDanny Banks

Chapter 14Molecular and nanoscale electronicsMichael C. Petty and C. Pearson

Appendix Summary of computer programs for analysis and design

1100_S01.fm Page 1 Friday, January 31, 2003 3:16 PM

section one

Introduction


1100_ch01.fm Page 3 Friday, January 31, 2003 3:54 PM

chapter one

Introduction to neuroprosthetics

Warren E. Finn and Peter G. LoPresti

Contents

1.1 Purpose of the handbook1.1.1 Why a handbook on neuroprosthetics?1.1.2 What this book hopes to accomplish

1.2 Evolution of neuroprosthetics1.2.1 Early experimentation and technologies1.2.2 Key tools arrive and first successes reported

1.2.2.1 Key tools1.2.2.2 First successes in neuroprosthetics

1.2.3 Rapid expansion1.3 A marriage of biology and engineering

1.3.1 How do neurons communicate with each other? 1.3.2 How does one communicate with neurons?1.3.3 How does one make an implant last?

1.4 Organization and contents of the book1.5 SummaryReferences

1.1 Purpose of the handbookSince 1990, the field of neuroprosthetics has grown at a tremendous rate.But, what exactly is meant by the term neuroprosthetics, and why areneuroprostheses of such consuming interest? For the purposes of thishandbook, a neuroprosthetic is a device or system that does one of thefollowing:


1. Replaces nerve function lost as a result of disease or injury. The


neuroprosthetic commonly acts as a bridge between functional ele-ments of the nervous system and nerves or muscles over whichcontrol has been lost. Examples include the peripheral nerve bridgesimplanted into the spinal cord, lumbar anterior-root stimulator im-plants to allow standing in paraplegics, and systems to restore handand upper limb movement in tetraplegics. The neuroprosthetic mayalso act as a bridge between the nervous system and a physicalprosthesis, as is the case in upper limb replacement.

2. Augments or replaces damaged and destroyed sensory input path-ways. The neuroprosthetic records and processes inputs from outsidethe body and transmits information to the sensory nerves for inter-pretation by the brain. Examples include the cochlear implant forrestoring hearing and an assortment of retinal and visual cortexprostheses for restoring vision.

A common component of all the systems in Figure 1.1 is the need tointeract directly with nerves. The system must either collect signals fromnerves or generate signals on nerves, or both. The interaction may be withindividual nerve cells and fibers or with nerve trunks containing hundredsto millions of axons. Just as important is the need to understand and speakthe language of the nervous system and understanding that the languagechanges as the signaling requirements change. For example, the auditoryand optic nerve systems have very different organizations, levels of signal-ing, and processing complexity as dictated by the different nature of theauditory and visual inputs. A neuroprosthesis, therefore, is a device or sys-tem that communicates with nerves to restore as much of the functionalityof the nervous system as possible.

1.1.1 Why a handbook on neuroprosthetics?

The rapidly expanding interest and research in neuroprosthetics over thelast decade paralleled the rapid increase in resources and literaturedevoted to the larger fields of bioengineering and biomedical engineering.A quick search of the Internet finds over 50 academic institutions withdepartments of bioengineering, many of which are less than a decade old,and many more with a bioengineering or biomedical engineering “empha-sis” within traditional departments such as chemistry, electrical engineer-ing, mechanical engineering, and biology. Professional publications thatpresent research in bioengineering continue to increase in number and insize. An excellent example of this is the IEEE Transactions on Systems, Manand Cybernetics, which was founded as a single entity in January of 1971,was split into two parts in 1996, and had to be split yet again into threeparts just two years later in 1998. New titles, such as the IEEE Transactionson NanoBioscience (due to be published in late 2002 or early 2003), reflectthe effect of emerging technologies on the practice of bioengineering and



biomedical engineering. Government spending and support, alwaysimportant in the sciences, has also been increasing, as evidenced by suchinitiatives as the German Federal Ministry of Education and Science andestablishment of the National Institute of Biomedical Imaging andBioengineering in 2000 at the National Institutes of Health in the UnitedStates. While the wealth of support, research, and literature is a goodthing, it can present even an experienced practitioner with the dauntingtask of assembling the basic knowledge required to design and implementan effective neuroprosthesis from widely spread resources. One reason forthe book, then, is to provide a point of consolidation for key informationthat aids practitioners in more effectively finding the techniques andinformation they need.

Figure 1.1 Diagram of human body showing many of the neuroprosthetic systemscurrently employed or in development.


Another reason for developing this book is the inherent interdisciplinary


nature of the neuroprosthetics field. Practitioners from such disparate fieldsas electrical engineering, mechanical engineering, mathematics, physics,computer science, physiology, neurology, pharmacology, and cellular biol-ogy, to name just a few, must work together in a cooperative environment,without the benefit of a common vocabulary, a common set of methods forapproaching problems, or a common set of analytical and experimental tools.Yet, methods and concepts from all of these areas are necessary to build aneffective neural prosthesis. We give some examples of this marriage of engi-neering and biology later in this chapter. The second reason for the book,therefore, is to provide a common point of reference to facilitate interactionsand understanding among practitioners from different backgrounds.

The final reason for developing this book is the potential impact that thedevelopment of neural prosthetics may have on society as a whole. Neuralprosthetics have the power to significantly extend the lifespan of a person andincrease the portion of that lifespan during which a person is an active andproductive member of society. Profound changes in workforce demographics,national healthcare systems, and the way in which people participate in societyin their later years are quite likely to follow as more people live longer andlead more active lives. Neural prosthetics have the potential to reverse, in totalor in part, the loss of function not related to aging, which may lessen thephysical and psychological impact of injury or disease. In addition, neuropros-thetics have the potential to extend the capabilities of the human body beyondits current limitations. Visual prostheses using semiconductor-based photore-ceptors, for example, could extend the visual experience into the infrared. Thepotential impact on social and political systems will likely be significant aswell, though this is beyond the scope of this book.

1.1.2 What this book hopes to accomplish

The purpose of this book is threefold. First, the book intends to combine themost commonly employed concepts, applications, and knowledge from themany disciplines associated with neuroprosthetics in a clear and instructiveway. Mathematical and modeling theories combine with examples of theiruse in existing and future neuroprostheses to clarify their usefulness anddemonstrate their limitations. Second, the book intends to provide examplesof neuroprosthetic systems at different stages in their development, from themore mature cochlear implant to the maturing area of upper-limb control tothe still unsettled world of visual prostheses. The book hopes that exploringthe varying developmental processes will provide guidance for those devel-oping other prostheses in regard to potential pitfalls, looming issues thatmust be solved, and successful strategies for solving difficult problems.Third, the book hopes to introduce key topics at a level that is useful to bothnew and practicing professionals working directly or indirectly with neuro-prosthesis projects. By providing an accessible common ground, the bookhopes to foster a more effective and productive collaborative environment.



1.2 Evolution of neuroprostheticsNeuroprosthetics, while currently a field generating a lot of interest andscholarly work, has reached its current state through a traceable evolutionarypath. Some of the key events and technologies that have driven this evolutionare noted on the timeline in Figure 1.2. As one can see from the timeline, along period of steady improvements in technology and groundbreakingexperiments has led to an explosion of new and more functional systems inthe last few years. In this section, we briefly discuss some key aspects of theevolution of neuroprosthetics.

1.2.1 Early experimentation and technologies

The field of neuroprosthetics has a long history, tracing its roots back to the18th century. Luigi Galvani observed at that time that a frog’s skeletal mus-cles contracted when in contact with both an anodic and a cathodic metal.Allesandro Volta, his contemporary, connected his newly discovered “bat-tery” to his ear and discovered that an aural sensation could be inducedelectrically. These early experiments were severely limited by the availabletechnologies, particularly in the equally fledgling area of electricity.

As technology and the fields of physiology and biology advanced, theexperiments of Galvani, Volta, and other pioneers were periodically revis-ited. It was not until 1934, however, that the first attempt at developingsomething like a modern prosthesis was successful. At that time, the firsttrue electronic hearing aid was developed, based on the work of Wever andBray.1 While admittedly crude, it was the first real indicator that meaningfulimprovement in sensory function by an electronic device was possible. How-ever, real breakthroughs would not be achieved until engineering and bio-logical methods improved to the point that communication on the cellularlevel was possible.

1.2.2 Key tools arrive and first successes reported

1.2.2.1 Key toolsIn the middle and late 20th century, researchers received several key toolsthat would facilitate the development of more successful neuroprostheses.The first of these tools, the transistor and its platform, the integrated circuit,evolved over a period from 1947 through 1963. The transistor provided ahost of capabilities in a package that would rapidly decrease in size. Currentscould be more precisely controlled and switched using small voltage acrossa metallurgical junction between positively doped (p-type) and negativelydoped (n-type) semiconductors, and this current could be made independentof the controlling circuit. The transistor was also capable of significantlyamplifying weak signals without the need for bulky and radiative trans-formers, which eventually reduced the size and cost of signal amplificationcircuitry required to “hear” neurons and nerves talking. The technology




uter milestones and progress in the development


Figure 1.2 Historical timeline showing the relationship between engineering and compof neuroprosthetic devices.


required to produce the MOSFET (metal-oxide semiconductor field effect


transistor) is a particularly important milestone, as this led to the develop-ment of charge-coupled device (CCD) cameras and variations capable ofincorporating neurons into the structure for either recording or stimulation.2,3

The development of integrated circuits (ICs) and silicon chips made thetechnology more manageable and simpler to use.

The seminal work of Hodgkin and Huxley appeared in 1952 and continuesto influence the way researchers envision the communication between elec-tronics and tissue to this very day (see Chapter three).4 Based on a series ofexperiments on squid axons, Hodgkin and Huxley developed a physiologicalmodel of neuron behavior at an unprecedented level of detail. In particular,the model illuminated the process of action-potential generation, detailing theroles of membrane potential and ionic currents. The work also describedunique methods for controlling and recording neural signaling. Presentresearchers use the information contained in the model to develop electricalmodels of neuron activity, select the proper methods for eliciting a desiredresponse, and deciphering the origin of measured signals (see Chapters three,four, and five for basics; also see discussions in Section five).

As the ability to communicate with neurons grew and the modelsdescribing the process became more sophisticated, a tool was required toallow scientists to control more complex experiments, to process electricalsignals more rapidly, and to isolate the contributions of individual neurons.The development of the microprocessor in 1971 and very-large-scale inte-gration (VLSI) in 1977 provided the necessary tools. The techniques devel-oped to construct VLSI circuits formed the foundation from which circuitminiaturization and micromachining of microelectromechanical systems(MEMS) and electrode arrays were developed (see Chapters six, seven, andthirteen, for example). Transistors and other electronic structures the size ofa neuron were now possible. The microprocessors provided researchers witha means to program and automate experimental procedures, data collection,and data processing. While specialized processor chips (such as digital signalprocessing [DSP] and analog-to-digital converter [ADC] chips) would notcome until later, the microprocessor still facilitated a dramatic increase inexperimental complexity and control with a correspondingly dramaticdecrease in execution time. As microprocessor technologies and architecturesmatured, experiments requiring simultaneous monitoring of multiple signalsbecame feasible. This led to new discoveries such as the concerted signalingin the retina that offers clues to the processing of visual images.5

The combination of affordable microprocessors and VLSI led to one ofthe most important tools at the disposal of modern researchers, the affordablepersonal computer. The first IBM personal computer appeared in 1981, andthe computer has since become a staple of every laboratory and researchcenter. Not only has the computer increased the degree of automation inexperimental work, but it has also made the processing and analysis of dataeasier. Complicated control algorithms for artificial limbs can be tested andmodeled easily. Images from microscopes and analog signals from recording


electrodes are captured directly into the computer, reducing the need for


such time-consuming activities such as scanning pictures, developing film,and data entry. Many modern analysis and processing tools for extractinginformation from images and data, such as the wavelet transform in Chaptereight, would be laborious or impossible without the capabilities of the mod-ern computer. Most importantly, many experiments and tests can be per-formed completely on the computer, without the need for biological subjects,in order to perfect designs and minimize hazards before live testing begins.More powerful programming languages and user-friendly interfaces havehelped to ensure the continued utility of this most useful tool.

Finally, the development of the scanning tunnel microscope has allowedresearchers to visually explore the world in which they operate. Combinedwith cell staining and other cell marking techniques, one is now able toobserve the growth of artificial neuronal networks, explore the impact ofimplants on cell pathology, and actually view the tiny microelectrodes andsimilar devices, just for starters. Assays such as these provide invaluableinformation on the behavior of nerves and neurons and their interactionwith foreign implants that improve the design of the next generation ofdevices and systems.

1.2.2.2 First successes in neuroprostheticsAs the tools noted above became available and more widespread, a numberof successful prosthetic systems were developed. The development of justtwo such systems are detailed below and in Figure 1.2. Auditory prosthetics,building on the early success with the hearing aid, were among the first tobenefit from the new technologies. The first cochlear implant was developedin 1957 by Djourno and Eyries1 and consisted of electrodes placed on theauditory nerve and stimulated at different pulse rates. By the mid-1970s,the cochlear implant had been refined to the point that clinical trials werebegun in the United States, and a bone-anchored hearing aid was madeavailable in Europe. By 1980, the cochlear implant became importantenough to warrant a U.S. Food and Drug Administration (US-FDA) Inves-tigative Device Exemption to clinically test a middle ear implantable hearingdevice (MEIHD), and by 1983 clinical trials had begun in Japan as well. Bythe early 1990s, the cochlear implant began to gain widespread commercialand public acceptance.

The other prosthetic systems developed at this time addressed losses inmotor function. The first motor prosthesis, targeted at foot-drop in hemiple-gics, was developed in 1961. Despite this early success, it was not until themid-1980s that clinical trials definitively proved that functional electricalstimulation (FES) of motor nerves and muscles was a valid approach. Thesetrials showed that FES could allow paraplegics to perform the actionsrequired to stand. By the mid-1990s, several versions of a neural prosthesisfor standing had been developed and approved for human trials, and sys-tems for biotic hands, upper-limb prostheses, and systems to treat urinaryincontinence had begun in earnest.


1.2.3 Rapid expansion


In the last five years alone, the field of neuroprosthetics has seen a rapidgrowth, both in the enabling technologies and in the number of biologicalsystems targeted by neuroprostheses. Again, Figure 1.1 provides a summaryof the more prominent prosthetic systems under development.

In 1998, the US-FDA approved the Finetech–Brindley bladder controllerfor commercial use, and the first totally integrated cochlear amplifier (TICA)was implanted in Europe. In 2000 alone, the US-FDA approved the firstmiddle ear implant, the auditory brainstem implant, and the Interstimimplant for bladder control for use in humans. Also, a fully implantablehearing aid, the Implex AG Hearing Technology from Germany, wasapproved for European use. Prostheses for restoration of vision began tomake significant progress with large-scale human trials of prostheses locatedin the visual cortex (1995), epiretinal space (1998), and subretinal space(2000). The results of the US-FDA-authorized subretinal trial were presentedin 2002 (see Chapter 11 for details). Implantation of the Abio Cor, a perma-nent, self-contained heart replacement proceeded in 2001, along with PhaseII studies on a totally implantable MEIHD. Also in 2001, the US-FDAapproved the first contactless middle-ear implant and the Handmaster sys-tem for restoring hand functionality.

A host of technologies and extensive research have built upon the earlierdeveloped tools to fuel this explosion in neuroprosthetics. The continuedminiaturization of all forms of electronics, from cameras to processors to theelectrodes themselves, has made it more feasible to communicate with largernumbers of nerves and neurons, therefore providing finer control over motorfunctions and finer sampling of sensory inputs to the ears and eyes. Com-munications technologies, both optical and electronic, have reduced the needfor control wires and transcutaneous electrical connections and are key toliberating the implant recipient from excessive external apparatus and pre-venting infection. Emerging technologies such as MEMS (see Chapter thir-teen), biomolecular electronics (Chapter fourteen), and artificially grownneuronal networks6,7 have provided the means to better understand neuralbehavior and engineer effective and long-lasting prostheses. Materials andmethods for reducing the rejection of the foreign prostheses by biologicaltissue have matured and are well known (see Section 12.3 of Chapter twelvefor a basic discussion). The impacts of fields such as nanotechnology andgenetic engineering have yet to be felt.

1.3 A marriage of biology and engineeringAs evidenced by the timeline in Figure 1.2 and discussions of the previoussection, the development and realization of neuroprosthetic devicesrequire the talents and knowledge of both the biologist and the engineer.Even basic experimentation, whether in the laboratory or in a theoreticalframework, cannot be performed without appropriate knowledge of


biological and engineering techniques and technologies. In this section we


discuss briefly some of the key issues that are central to the successfuldevelopment of a neuroprosthesis and that require a combined biologi-cal/engineering solution.

1.3.1 How do neurons communicate with each other?

Before one can begin talking to nerves and neurons, one must first determinehow they communicate with one another and determine how informationis encoded within the communication. The challenge, then, is to monitor thesignals transmitted by one or more neurons to a tightly controlled naturalstimulus and correlate features of the signals with information contained inthe stimulus.

Several methods have been devised to record from nerves and neurons,based on biological knowledge of how nerves conduct signals. Most nervescommunicate via action potentials, a complex signal generated by an intri-cate coordination of ion movements across neuronal membranes (see Chap-ters two and three) and controlled by voltage potentials across the cellmembrane. Recording devices must therefore tap or intercept voltages andionic currents, and transform them into electrical signals suitable for pro-cessing. While the concept is relatively simple, implementing such a deviceis complicated by the millimeter to micrometer scale of most neurons andthe small changes (millivolts or lower) in membrane potentials typicallyencountered. Material scientists are required to develop devices small andreliable enough to interact with a neuron. Devices such as the cuff electrodeand suction electrode8 measure a compound signal from the entire nerve,while single-wire electrodes and electrode arrays (see Chapter seven) aimto record from one or a small population of neurons, respectively. Electricalengineering techniques are required to extract the neural signal from biolog-ical and external noise sources and amplify them to manageable levels forprocessing. Solutions are found in physical differential amplifier-based headstages and in analog filters and amplifiers, in addition to the software ormicroprocessor-based solutions that use sampled representations of therecorded signal.

Correlating features of the neural signal with the original stimulusrequires knowledge of the biological system under study and powerfulanalysis tools to examine the myriad of possibilities. In the ear, for example,the axons that extend from the cochlea to make up the auditory nerve areknown to respond to specific sonic frequency ranges distributed along thelength of the cochlea (see Chapter ten). This knowledge of how the neuralsystem functions provides important clues to interpreting signals from dif-ferent sections of the auditory nerve. It is also generally agreed upon thatthe frequency, timing, and duration of action potentials generated by a givenneuron carry a significant amount of information. Analysis tools must there-fore be able to track amplitude and frequency as a function of time. Wavelettheory has proven to be a useful tool, though not the only one, in addressing


this challenge. Artificial neural networks and other sophisticated statistical


analysis programs provide the computational power necessary to siftthrough a variety of possibilities and to arrive at the most likely relationshipsbetween stimulus and response (see Chapter twelve for example).

1.3.2 How does one communicate with neurons?

While listening in on the conversation between neurons is a challenge, learn-ing how to inject our thoughts into the conversation is also difficult. We mustmanipulate voltages or inject currents to make ourselves heard withoutdamaging the cells or their surroundings and ensuring that our messagereaches the intended cell or group of cells. The challenge then is to find themost effective and safest way to communicate with neurons.

Extensive research continues to focus on how to best communicate withcells. While impaling a cell with an electrode is the most direct approach,the cell inevitably dies from the wound, and the approach is not practicalfor a functional neuroprosthesis. Many of the methods used to listen in oncells also function well as signal transmitters. Regardless of the methodemployed, the key issues that must be addressed are the amplitude of thestimulating signal (voltage or current), the duration and polarity of thesignal, and the spatial selectivity. To be successful, the biologist must inves-tigate how the natural processes of a cell are altered by a foreign stimulusand must determine the limits of this response before damage occurs. Forexample, a biphasic (two-polarity), charge-balanced signal best replicates thenatural ebb and flow of ionic currents when a current stimulus is used (seeChapter four). Engineers, material scientists, and physicists must then findthe best way to generate such a signal and deliver it to the cell. Sophisticatedmodeling techniques, such as those described in Chapters five and seven,estimate the voltage or current generated by competing electrode designs asa function of time and space within adjacent tissue. By adjusting the prop-erties (surface area, geometry, and conductivity), researchers attempt to tar-get specific cell groups with a sufficiently large stimulus. Different implantmaterials and architectures influence the electrical power required to delivera desired density of charge to the cell, which in turn affects the electricalefficiency and heat generation of the implant. As the available technologycontinues to evolve, researchers continue to refine their techniques.

In addition to creating a safe and effective electrical connection with thecell, the implant must not physically endanger the cell and its surroundings.Many implant materials, such as semiconductors and most metals, are poi-sonous to the human body. Insulating materials such as silicone prevent anyinteraction between the poisons and the tissue without acting as a barrier tothe electrical signals. An implant must not cause tearing or other physicaldamage to the tissue at the point of connection. The human body experiencessignificant amounts of movement and jarring impacts in even a normal day,and the implant must move in concert with the surrounding tissue to avoidinjury. Implants must also allow the exchange of nutrients and waste to


proceed naturally so that the surrounding tissue remains healthy. If the cells


that communicate with the implant perish, the implant becomes ineffective.These issues are treated with greater detail in Chapter eleven.

1.3.3 How does one make an implant last?

Interrelated with the first two issues is the problem of making a devicepalatable, or at least invisible, to the natural defense mechanisms of thehuman body. In addition, the natural environment within the body is detri-mental to the long-term integrity and survivability of implant materials. Thechallenge here is to design a system that will operate successfully forextended periods in a hostile environment.

One of the keys to a successful chronic implantation is a judicious choiceof materials. Fortunately for researchers, a number of materials have beenidentified as safe for use in the human body and most likely to survive thebiological environment (see Chapter eleven). Safe materials exist to performall the basic functions required within a neural prosthesis, from carryingelectrical charge to electrical and physical insulation. Engineered biologicalmaterials may add to this list and provide greater functionality and surviv-ability of implants.9,10 A design that incorporates these materials, therefore,is more likely to survive chronic implantation.

Another key issue is that of wound healing. All implants, regardless ofdesign, create a wound of one sort or another upon placement in the body.Small wounds can continually occur as the implant sight moves and flexeswith movement of the body. How the body reacts to these wounds deter-mines the long-term effectiveness of the implant. For example, prosthesesimplanted on the surface of the retina are encapsulated by fibrous growthsdrawn from the vitreous humor if care is not taken during the implantingprocedure. This encapsulation increases the physical distance between theimplant and the targeted cells. Because the emitted electric field strengthdecreases inversely with distance, a point is rapidly reached where the fieldamplitude is insufficient to stimulate the targeted cell, and the prosthesisbecome ineffective. Engineers must work with biologists and surgeons toensure that the implanting procedure does not adversely affect the lifetimeof the implant.

1.4 Organization and contents of the bookThe book is organized into five main sections beyond this introductorysection. In Section II, the basic elements of neural behavior are describedand modeled. Chapter two addresses the fundamental processes of neuronactivity, including neuronal excitability and its regulation by membrane ionchannels. These processes are modeled mathematically in Chapter three. Themodels presented in Chapter three are some of the basic tools for choosingthe placement and type of stimulating electrodes and estimating the requiredstimulus amplitude to elicit a desired neural response.


Section III provides the bridge between the biological and engineering


worlds through discussion of the interface between electronics and neuraltissue. Chapter four describes the basic methods and findings related tostimulating neural activity with electrical signals. Here, one will find whatmethods are best for targeting specific cells or cell components (axon, den-drite, or soma) for activation and how to minimize cell damage duringchronic stimulation. Chapter five details specific models for better predictingthe behavior of electrodes within a biological medium, complementing thediscussion in Chapter two. In Chapters six and seven, semiconductor-basedsystems for communicating with implants, integration of signal processingand power supplies with the implant, and procedures for recording activityfrom neurons are presented.

Section IV discusses some of the more common techniques required tointerpret, process, and utilize neural signals for neuroprosthetic systems.Chapter eight details the basics and proper use of the powerful wavelettransform for extracting key informational components from neural signals.In Chapter nine, the problem of using neural signals in neuroprostheticdevice design is addressed.

A description of several different neuroprosthetic systems and theirdevelopment is the subject of Section V. Chapter ten details the state of therelatively more mature field of otological implants for hearing rehabilitation,while Chapter eleven details the ongoing development and challenges ofvisual prostheses. Chapter twelve provides an excellent overview of themotor prosthesis field, which has challenges very different from thoseencountered with sensory prostheses.

Finally, in Section VI, two technologies are discussed that will have asignificant impact on the future of neural prosthetics. In Chapter thirteen,the continuing progress in the field of microelectronics is discussed, includ-ing the application of MEMS devices. Chapter fourteen summarizes theexciting new area of biomolecular electronics, where biological entitiesprovide the bridge between electronics and neurons. Further informationuseful to the neuroprosthetics practitioner is included in the appendicesin Section VII.

1.5 SummaryThe field of neuroprosthetics is an exciting and challenging one. Practitionersfrom diverse backgrounds must communicate and innovate together todevelop effective and practical systems and devices. It is our hope that thisbook will serve as a common ground on which these practitioners can cometogether, forge understandings, and acquire the basic knowledge theyrequire. From better communication, better interaction, and enhanced aware-ness of concepts and methods can spring innovative and more sophisticatedideas and implementations that will not only benefit the prosthesis recipient,but will also help to further unlock the mysterious and wondrous workingsof the human body and mind.



References1. Miyamoto, R., Cochlear implants, Otolaryngol. Clin. North Am., 28, 87–294,

1995.2. Fromherz, P. and Stett, A., Silicon-neuron junction: capacitive stimulation of

an individual neuron on a silicon chip, Phys. Rev. Lett., 75, 1670–1673, 1995.3. Meister, M., Pine, J., and Baylor, D.A., Multi-neuronal signals from the retina:

acquisition and analysis, J. Neurosci. Meth., 51, 95–106, 1994.4. Hodgkin, A.L. and Huxley, A.F., A quantitative description of membrane

current and its application to conduction and excitation in nerve, J. Physiol.,117, 500–544, 1952.

5. Meister, M., Lagrado, L., and Baylor, D.A., Concerted signaling by retinalganglion cells, Science, 35, 1535, 1995.

6. Corey, J.M., Wheeler, B.C., Brewer, G.J., Compliance of hippocampal neuronsto patterned substrate networks, J. Neurosci. Res., 30, 300–307, 1991.

7. Branch, D.W., Wheeler, B.C., Brewer, G.J., and Leckhand, D., Long-term main-tenance of patterns of hippocampal pyramidal cells on substrates of polyeth-ylene glycol and microstamped polylysine, IEEE Trans. Biomed. Eng., 47,290–300, 2000.

8. Stys, P.K., Ransom, B.R., and Waxman, S.G., Compound action potential ofnerve recorded by suction electrode: a theoretical and experimental analysis,Brain Res., 546, 18–32, 1991.

9. Zhong, Y., Yu, X., Gilbert, R., and Bellamkonda, R.V., Stabilizing elec-trode–host interfaces: a tissue engineering approach, J. Rehabil. Res. Develop.,38(6), 2001.

10. Haipeng, G., Yinghui, Z., Jianchun, L., Yandao, G., Nanming, Z., and Xiufang,Z., Studies on nerve cell affinity of chitosan-derived materials, J. Biomed. Mater.Res., 52(2), 285–295, 2000.



section two

Neurons and neuron modeling


chapter two

Neuronal excitability: membrane ion channels

Steven Barnes

Contents

2.1 Introduction2.2 The balance of passive and active properties

of the neuronal membrane2.3 Active membranes overcome temporal and spatial

degradations caused by passive properties2.4 Gating and permeation: the facts of life for ion channels2.5 Families of channels for Na+, K+, and Ca2+

2.6 Molecular properties of voltage gated Na channels2.7 The voltage-gated K channels show great structural diversity2.8 The family of voltage gated Ca channels2.9 ConclusionReferences

2.1 IntroductionThe purpose of this chapter is to provide an understanding of how ionchannels produce membrane excitability. The neuroprosthetic theories anddevices described in this book depend to varying degrees on the excitabilityof neural membranes and on the function of many different types of voltage-gated ion channels. The overall problem is that neuroprosthetic devices, suchas retinal implants, cochlear implants, deep brain stimulators, and motorprostheses, achieve their effect by selectively exciting a neuron or nerve. Inthe retina, it can be the ganglion cells or the bipolar cells; in the ear, the spiralganglion neurons; or for motor prostheses, near the nerve–muscle junction.

1100_ch02.fm Page 21 Thursday, November 7, 2002 3:59 PM


In each example, an interface exists between a stimulating electrode and aneuronal membrane, and once stimulation of the neural membrane occursthe cell physiology itself must be relied upon to carry the signal to itstermination points.

This chapter provides explanations of the voltage-dependent gating andthe permeation properties of several types of ion channels in neuronal mem-branes. The emphasis here is on voltage-gated ion channels, but the involve-ment of other channel types in neuronal excitability should not be excluded.The first focus is on the passive electrical properties of the typical neuronalsomatic membrane and axon to recognize the important role that active electricalproperties play to overcome the restrictions on electrical signaling that thepassive properties confer. The ion channels featured most prominently in thisbalance are voltage-gated sodium channels (Na channels) and numerous typesof voltage-gated potassium channels (K channels). An additional class of volt-age-gated channel types that are considered are calcium channels (Ca channels),which are active in virtually every cell type, neuronal or otherwise, and playimportant roles in cell signaling via Ca2+ as an intracellular messenger.

2.2 The balance of passive and active properties of the neuronal membrane

Signaling via membrane potential (voltage) changes in cells lacking activeconductances is dramatically limited by the passive properties of the cells.Passive properties arise for the most part from the membrane capacitanceafforded by the lipid membrane and the resistance of the intra- and extra-cellular fluids, as well as the resistance of the membrane itself. Lipid bilayersare of extremely high resistance, so this component of the passive repertoireis not considered a restricting factor. However, the combination of capaci-tance and resistance gives rise to spatial and temporal filtering of voltagesignals. That is, the distance that a signal can propagate without seriousdecrement is limited, and the fidelity of a signal in time is markedly reduced,typical of low pass (RC) filter characteristics.

The extremely fine neuritic process that a spinal motor neuron employsto connect to its target muscles, or the axon projecting from the ganglion cellsof the retina to the thalamus of the brain, must in a reliable manner carry high-frequency information of up to 1000 Hz over distances ranging from a fewcentimeters to a meter. Cable theory adequately approximates the distancethat a sustained membrane potential change can be detected along an axon:V(x) = V(0)exp(–x/

λ) in one dimension, with the length constant given by

λ= (rm/(ri + ro))1/2. For an axon of 1

µm diameter, the length constant is on theorder of a few hundred micrometers. A single exponential equation alsoapproximates the time dependence of the rise and fall of the signal as V(t) =V(0)exp(–t/

τ), with the time constant given by

τ = rmcm. Changes in voltageare thus slowed dramatically by cells, where typical values for the time con-stant are on the order of 1 to 100 msec. (See Figure 2.1.)



Considering the spatial and temporal limitations imposed by a passivemembrane, propagation of the electrical signals is poor. The failure of an axonalprocess to carry information over a distance of more than a few hundredmicrometers necessitated the appearance of ion channels, which, together withionic gradient-generating and energy-requiring ion pumps and exchangers,provide the means for rapid and reliable regenerative electrical signaling.

Figure 2.1 Decay in time and space of an electrical signal in an axon due to passiveproperties of the membrane. (A) The segmental resistances (ro, resistance of outsidemedium; rm, resistance of membrane; ri, internal resistance within the axon) andcapacitance (cm, membrane capacitance) involved in the passive propagation of adepolarization. In the diagram, current (I) is injected at one end with a microelectrode,and membrane voltage at three positions (V1, V2, and V3) is measured along the axon.(B) Voltage signal at the three positions. Close to the site of current injection, voltageV1 is the largest and is only rounded slightly by the RC filtering properties of thepassive membrane. With increasing distance from the site of current injection, voltagesignals V2 and V3 are smaller in amplitude and reflect a marked slowing in rise time.(C) Plot of membrane voltage against distance along the axon. The decay of theamplitude of the signal follows a single exponential function characterized by thelength constant, λ, the distance taken to reduce the signal to 37% of its original value.(From Hall, Z.W., An Introduction to Molecular Neurobiology, Sinauer Associates, Sun-derland, MA, 1992. With permission.)



2.3 Active membranes overcome temporal and spatial degradations caused by passive properties

Several different types of ion pumps and exchangers produce ionic gradi-ents between the inside and outside of cells. Energy-requiring pumps, suchas the sodium–potassium ATPase, which uses one ATP molecule to drivethree Na+ ions out and two K+ ions in, are the primary source of the ionicgradient almost always found in neurons. The high sodium concentrationoutside the cell (for mammals, this is about 145 mM) and low concentrationinside the cell (below 10 mM) produces an inward-driving force on Na+

ions. The high K+ concentration inside the cell (around 150 mM) and thelow concentration outside (usually about 3 mM) produces an outward-driving force on K+ ions. A calcium pump called the Ca2+-ATPase uses ATPto extrude Ca2+ ions from the inside of cells and produces extremely largeconcentration gradients for this ion with external concentrations of 2 to 3mM and internal concentrations around 10 nM. The Ca2+-ATPase is assistedby a class of exchangers that utilize the inward Na+ gradient to move Ca2+

out. Additional pumps and exchangers move Cl–, bicarbonate (HCO3–), and

other ions across the cell membrane.With the gradients established, gating of ion channels can produce a

remarkable diversity of electrical signals. The most frequently encounteredelectrical signal in nerve cells and muscles is the action potential, whichrepresents a brief, transient, self-regenerating depolarization. The stereotyp-ical action potential has rising, repolarizing, and after-hyperpolarizingphases that were described nearly a century ago. From a resting potentialof near –60 mV (this value varies considerably between cells), an externalstimulus (depolarizing influence) may bring the membrane to threshold(near –40 mV; again, the value will vary between cells). Once threshold isreached, the cell fires an all-or-none, regenerative action potential whosepeak amplitude may reach +40 mV. The repolarization phase returns mem-brane potential to a value near the original resting potential, often with abrief period of “after hyperpolarization” during which the membrane poten-tial is made more negative than the resting value. The time course of anaction potential varies, with most mammalian neurons firing spikes havinga half-width of well under 1 msec.

Physiologists studying the problem of cellular excitability were awareof the different ionic compositions of the intra- and extracellular media andconcluded that ionic conductances must vary over time to account for themembrane polarizations seen during the action potential. Hodgkin andHuxley6 voltage clamped the giant axon in squid and showed that anincreased sodium conductance was responsible for the rise of the actionpotential and that increased potassium conductance was responsible for therepolarizing phase. Figure 2.2 shows the time courses of the transient sodiumconductance increase (gNa) and, following with a short but essential delay,the transient potassium conductance increase (gK).



Hodgkin and Huxley6 modeled total membrane current recordedunder voltage clamp. Their recording of the membrane currents activatedover the same voltages at which action potentials occur showed a fast,transient depolarizing current followed by a delayed and sustained hyper-polarizing current (Figure 2.3). Their observations gave rise to concept ofion channels long before channels could be identified as individual mem-brane proteins. Before the single-channel recording period was ushered inwith the advent of patch clamping, physiologists and biophysicists couldonly record “macroscopic” or “whole cell” currents, which reflect theensemble of thousands of individual channels undergoing their gating andpermeation activities together.

Thousands of individual ion channels are responsible for the membraneconductance changes shown in these classical experiments. Ion channelactivity was inferred from biophysical experiments analyzing the noisepresent in macroscopic ionic currents, and it was clear that highly selectivetoxins could block separable components of the ionic currents in cells, indi-cating that channels selective for different types of ions were present, but itwas only with the patch clamp technique that single channel currents acrossthe membrane of cells could be definitively identified.

The electrical signals in many neurons expand on the theme of actionpotential production, with variations in the rate, temporal pattern, individualspike width, magnitude, and after-potentials. Other electrical signaturesinvolve the hyperpolarizing responses of cells. Generally, hyperpolarizationis opposed by inwardly rectifying K+ channels, although in many cases this

Figure 2.2 Time course of ionic conductances underlying the action potential. TheNobel-Prize-winning work of Hodgkin and Huxley identified transient increases inmembrane conductance to Na+ and K+ ions that followed very specific temporal pat-terns. The initial increase in Na+ conductance allows the cell to depolarize toward theNa+ equilibrium potential (ENa), an action that activates an additional increase in Na+

conductance in a positive feedback loop. The delayed activation of an increase in K+

conductance, together with inactivation of the Na+ conductance, causes the cell tohyperpolarize toward the K+ equilibrium potential (EK), a value negative to the normalresting potential of the cell. The long after-hyperpolarization can be seen to arise fromthe prolonged increase in K+ conductance. (From Hall, Z.W., An Introduction to MolecularNeurobiology, Sinauer Associates, Sunderland, MA, 1992. With permission.)



opposition occurs over different time scales with nonselective cation chan-nels activated by hyperpolarization or Cl– channels that produce inwardrectification at extreme membrane potentials.

2.4 Gating and permeation: the facts of life for ion channelsBiophysical descriptions of ion channel function generally fall within twocategories: gating and permeation. Gating describes the temporal depen-dence of the opening and closing of a channel and the probability of findinga channel in an open or closed state as a function of membrane potential orthe presence of a drug or neurotransmitter. Permeation describes the con-ductive properties of a channel in terms of its selectivity for specific ions,the rate at which ions can pass through the channel (which determines themaximum amplitude of the single channel current), and the effects and

Figure 2.3 Voltage-clamped sodium and potassium currents in an axon. Under volt-age clamp, membrane potential can be carefully and rapidly controlled to show thetime dependence and degree of activation of voltage-dependent ionic conductances.In this recording, membrane potential is stepped from a holding potential of –60 mV(chosen to mimic the resting potential) to new values between –10 and +90 mV insteps of 10 mV. A downward (or “inward”) current trajectory is recorded first atabout –10 mV that represents activation of the Na+ conductance, accounted for bythe ensemble activity of thousands of Na channels. At more depolarized values ofmembrane potential, such as +90 mV, upward (or “outward”) current is recorded.This is carried by K+ ions through the K+ conductance, again reflecting the activityof thousands of single K channels. (From Armstrong, C.M., J. General Physiol., 54,553–575, 1966, Rockefellar Press. With permission.)

1100_ch02.fm Page 26 Monday, November 11, 2002 10:26 AM


mechanisms of pore-blocking drugs and chemicals. Permeation and gatingproperties overlap in a few areas — for example, where an interaction of thepermeating ion with the channel pore can affect the time spent in the openstate of the channel.

Most ion channels have been described with gating schemes that followthe general pattern shown in Figure 2.4. The channel makes transitionsbetween closed, open, and inactivated states. In the case of many channels(some are described later in this chapter), no inactivated state occurs ortransitions to it are very slow or rarely made. Although the channel isnonconducting in both closed and inactivated states, inactivation is a specific,well-defined state from which transitions back to the open state are incred-ibly rare or nonexistent. The state transitions can be characterized withmembrane-potential-dependent forward and reverse rate constants that ulti-mately describe the time dependence of channel activation (transitions fromthe closed to open states), inactivation (transitions to the inactivated state,which are generally voltage independent), and deactivation (transitions fromthe open back to the closed state). Smooth S-shaped curves plot the proba-bility of a channel being open against membrane potential (activation curve).

Figure 2.4 State diagram of a voltage-gated ion channel and its relation to channelactivity. Most channels follow this scheme of closed–open–inactivated states, but withnumerous variations. In many channels, the inactivated state is not present; in others,multiple open and closed states exist, resulting in complex kinetics in the transitionsbetween opening and closing. In the classic view, the channel has at least two gates:(1) an activation gate that is sensitive to membrane potential and that allows the channelto make transitions from closed to open states, and (2) an inactivation gate on thecytoplasmic surface of the protein that renders the channel nonconducting in a voltage-independent manner following activation. The single-channel recording shows thechannel in the closed state, then in the open state, and then after a step back to theclosed state and a second opening, a transition to the inactivated state from which thechannel does not return until sufficiently hyperpolarized (not shown). (From Ashcroft,F.M., Ion Channels and Disease, Academic Press, San Diego, CA, 2000. With permission.)



While it was clear even in the early studies of ion channel gating thatthe channel must be capable of sensing the transmembrane potential, themechanism responsible for this only became evident after ion channelswere cloned and their molecular structure described. Hypotheses aboutwhich part of the protein had what function could then be tested onchannels engineered to contain specifically substituted amino acid resi-dues using the technique of site-directed mutation. It was shown thatone of the putative transmembrane (TM)-spanning regions, one with asurprisingly large number of positively charged amino acids and calledthe S4, was responsible for the voltage dependence of the voltage-depen-dent steps in activation. The positively charged residues could bechanged to residues with neutral side chains, with the result being thatthe channel no longer gated over the same range of membrane potentials.It has been hypothesized, but not proven, that the positive residues onthe S4 TM segment enable the helix to couple membrane potentialchanges to conformational changes in the channel protein, so as to favoran open conducting state with more depolarized values of membranevoltage (Figure 2.5).

From the biophysicists’ perspective, an ion channel required specificdomains to carry out the various known functions that had been carefullydescribed experimentally. Hille’s depiction of an ion channel protein (Figure2.6) carefully mapped out these domains in the era just before a thoroughdepiction based on molecular studies became available.

Figure 2.5 Catterall’s twisting helix model for the translation of membrane depo-larization into channel opening. Site-directed mutation studies identified the S4transmembrane helix region of the protein as that responsible for voltage sensing.Positively charged side groups of residues match up with negative charges onother helical segments, but the strength of the electric field within the membraneexerts force on the S4 helix and can alter which pairs of charges match up. Aratcheting twist of S4 produces a conformational change in the protein, allowingthe channel to open. Several other models have been proposed, related to thistheme. (From Ashcroft, F.M., Ion Channels and Disease, Academic Press, San Diego,CA, 2000. With permission.)



2.5 Families of channels for Na+, K+, and Ca2+

The explosion of molecular studies on ion channels that has occurred overthe past decade has provided a rich description of the structural basis ofthese proteins. Not only have diverse, seemingly unrelated sequences beenidentified in a few cases as subserving similar functions between someclasses of channels, but a provocative theme has also emerged stronglyindicating the evolutionary roots of the channels within and between classes.

Figure 2.7 provides the current perspective on the three principle classesof voltage-gated channels. The Na channels are all built around the themeshown in the figure: an

α subunit containing six transmembrane-spanninghelices repeating in each of four motifs. Two subunits can accompany thechannel, and these subunits in general may have roles in modulating theactivity of the channel. Figure 2.7B shows the structure of the

α1 subunits ofvoltage-gated Ca channels; this principle subunit is very similar in overallform with the

α subunit of the Na channel. The

α1 subunit can by itself forma complete channel, but accessory subunits (

β2,

γ, and

δ) bind to the channel

Figure 2.6 Hille’s pre-molecular depiction of the working parts of a voltage-gatedchannel. The protein sits in the membrane so as to produce a ring structure to permita central ion-permeable pore. The activation gate, shown here literally as a hingedstructure, is coupled to the voltage sensor of the channel. To permit permeation byspecific types of ions, the channel has a selectivity filter, a narrow region where theions bind transiently on their way through the channel. In this depiction, a cytoplas-mic region of the channel is poised to produce an inactivating block of the channelonce the activation gate swings clear. Ion channels are, like other proteins, glycosy-lated, and they are tethered to the cells cytoskeleton via anchor proteins. (From Hille,B., Ion Channels of Excitable Membranes, Sinauer Associates, Sunderland, MA, 2001.With permission.)



Figure 2.7 The principle subunits of voltage-gated ion channels. (A) Na chan-nels have a single

α subunit that is structured with four repeats of a six trans-membrane helix motif. The four motifs are number I to IV, and the six helicesin each motif are designated S1 to S6. Each S4 helix contains positively chargedamino acid residues and comprises the channel’s voltage sensor. The inactiva-tion mechanism is located on the cytoplasmic linker between motifs III and IV.The Na channel pore is constructed from the linking regions found between S5and S6 in each of the four motifs. Na channels are often composed of additional

β1 and

β2 subunits. (B) Voltage-gated Ca channels are structurally very similarto Na channels, with the same 6TM helix motif repeated four times. The S4voltage sensors are similar, as are the P-loop linkers between the S5 and S6helices of each motif. Inactivation of Ca channels is not the same, and severalregions of the channels and quite different mechanisms are responsible for thevarying degrees of inactivation seen in Ca channels. Ca channels almost alwayshave a diversity of accessory subunits as shown. (C) The simpler K channelwith its single-motif S1–S6 subunit. The same S4 voltage sensor region and P-loop pore domain between S5 and S6 are present. Inactivation of some voltage-gated K channels is mediated by the N-terminus region, where a ball of proteintethered to the S1 helix has been shown to literally plug the open channel. Kchannels are formed from four of these subunits, meaning that, with the diver-sity of subunits, heteromeric channels are found and channel properties arediverse. (From Hille, B., Ion Channels of Excitable Membranes, Sinauer Associates,Sunderland, MA, 2001. With permission.)

1100_ch02.fm Page 30 T

hursday, Novem

ber 7, 2002 3:59 PM


and are known to modify properties of the channel such as its interactionwith neuromodulators and its inactivation kinetics. It is the K channel thatappears to stand out because it is composed of four individual

α subunits,each one resembling the 6TM motif of the Na and Ca channels.

2.6 Molecular properties of voltage gated Na channelsThe Na channel of squid giant axon, whose kinetics were described byHodgkin and Huxley,6 certainly is archetypal for the subsequently discov-ered Na channels in a variety of other cell types. The hallmarks of Nachannel gating are the rapid transition from closed to open states withmembrane depolarization, and the subsequent rapid inactivation. Single-channel studies of Na channels brought forth a wealth of informationabout these transitions. With single-channel recordings, such as thoseshown in Figure 2.8, it was possible to show that channel activationoccurred over a relatively broad temporal window and that in nearly everycase inactivation was a voltage-independent, rapid, and essentially auto-matic process.

Inactivation of the Na channel is caused by a small structure composedof several amino acid residues on the cytosolic loop linking repeated motifsIII and IV. Shortly after the voltage-dependent activation of the Na channel(at 37°C, the process occurs in approximately 100

µsec), the hinged lidoccludes the open pore of the channel and essentially blocks the channel tofurther permeation.

Permeation of Na channels is determined by amino acid residues of theP-loop, a pore-forming domain present in each of the four motifs that linksthe TM5 and TM6 parts of each motif. Lining the pore of the Na channel arenegatively charged residues in two of the four P-loops, and these are essentialfor binding Na+ ions and conferring upon the channel selectivity for Na+.4

Sequence analysis of the nine expressed human Na channels showsa very high degree of amino acid identity between the proteins. Thedendrogram shown in Figure 2.9 emphasizes the similarities amonghuman Na channels, reflecting the relatively recent evolutionary diversityin this class of channel, as amino acid substitutions are a measure ofevolutionary distance.

2.7 The voltage-gated K channels show great structural diversity

Elucidation of the common structures of ion channels showed that Kchannels have the simplest subunit structure. Each subunit of a Shaker-type K channel has six transmembrane spanning helices. As describedabove, the S4 regions contain the voltage sensor, and the loops betweenS5 and S6, called the P-loops, fold inwardly and provide the pore of thechannel. Site-directed mutagenesis has identified which amino acid res-



idues of the P-loop are critical for permeation of K+ and block of thechannel by chemicals such tetraethylammonium (TEA). At the N-termi-nus of the 6TM subunit is a cytoplasmic domain shown to be responsiblefor fast, so-called N-type, inactivation.

Potassium channels activate more slowly than Na channels, and fromthe beginning were termed “delayed rectifier” channels, owing to this delayin activation and their current-voltage behavior, which has the tendencyto oppose strong depolarization (e.g., to give the membrane outwardly

Figure 2.8 Unitary Na channel currents account for the fast transient Na+ conduc-tance increase of the action potential. Panel (A) shows a single channel recording ofa 19-pS Na channel presented with many depolarizations. Each time the membraneis depolarized, the Na channel undergoes its transition from closed to open and thento inactivated states. With each depolarization, this process occurs at a slightly dif-ferent time due to the stochastic nature of channel gating. When the single openingsare gathered in an ensemble average, as shown in panel (B), the current waveformresembles macroscopic recordings made from intact axons. This patch actually con-tains several Na channels, accounting for the double opening in the uppermostexample. (Kindly provided by J.B. Patlak.)



rectifying properties). Figure 2.10 shows single K channel activity inresponse to a depolarization. The striking difference of these records, com-pared to those from the Na channel in Figure 2.8 (other than that the single-channel current is directed in the upward direction reflecting the outwardflow of positive K+ ions) is that the single channel opening are very longlived, for the most part lasting the duration of the depolarizing step. Theensemble current shown in panel B gives a clear perspective of the lack ofinactivation in the case of the K channel.

The single channel and ensemble averaged currents shown for Na andK channels in Figures 2.8 and 2.10 readily account for the macroscopiccurrents underlying the action potential shown in Figure 2.3. During theaction potential, the depolarizing phase is terminated by the combination ofNa channel inactivation and K channel activation. The repolarizing phaseof the action potential terminates, not because the K channels inactivate, butbecause just as the membrane becomes hyperpolarized the channels close,as shown in Figure 2.10.

Not all voltage-gated K channels follow the prototypical six-transmem-brane-spanning helix structure. As Figure 2.11 shows, several classes of Kchannels utilize subunits with two or four transmembrane-spanning helicesinstead of the 6TM subunit. Some of these are inwardly rectifying K channels,channels that oppose hyperpolarization of the membrane and contribute insome cases to the resting membrane potential.

Figure 2.9 Relationships between nine different human Na channels and their chro-mosomal locations. The dendrogram shows the percent identity of amino acids onthe various forms of the human Na channels. Channels having a high degree ofamino acid identity are considered to be recent evolutionary ramifications. Thenames of the channel proteins are shown with the corresponding human gene.Additional Na channel genes are known to exist but have not been expressed.(From Goldin, A.L. et al., Neuron, 28, 365–368, 2000. With permission.)



The list of channels provided in Figure 2.11 includes many not identifiedin humans. The principle class of voltage-gated K channels is that designatedby KV1 to 9. KV1.1, the Shaker channel, is the classic inactivating A-type Kchannel. It has roles in action potential repolarization, in some cases offeringcontrol of ramp depolarization of the membrane leading to threshold. Itsname comes from its discovery in a fruit fly mutation that shakes its legs

Figure 2.10 Unitary K channel currents account for the delayed and sustained K+

conductance increase of the action potential. (A) Single-channel recordings of Kchannels were recorded in a squid giant axon. A single channel with 20-pSconductance opened for most of the duration of each applied voltage step. Theopenings first occur with a much longer delay than the Na channel openings seenin Figure 2.9. The channel often makes transitions back to the closed state butthen reopens. When membrane potential is returned to a negative value, wherethe channel has a low probability of being in the open state, the channel alwayscloses rapidly. (B) The ensemble average shows a smooth activation time courseand a non-inactivating steady-state value during the step to +50 mV. This trajec-tory agrees well with the macroscopic K+ currents shown in Figure 2.4. (Withpermission from Francisco Bezanilla [see Reference 5].)



when anesthetized. The other KV channels offer unique channel kinetics andsingle-channel conductances.

Figure 2.11 K channels exist in many forms. K channels come not only as 6TMsubunits, but as 2TM and 4TM subunits as well. The 2TM subunits include inwardrectifier K channels and ATP-sensitive channels. Of the 6TM subunits, the largestsubfamily is the Kv group, including the sustained and rapidly inactivating K chan-nels. Ca2+-activated K channels include large and small conductance channels. Al-though not specifically K channels, the figure includes the nonselective cationchannels HCN and CNG, which are gated by hyperpolarization and cyclic nucle-otides, respectively. (From Hille, B., Ion Channels of Excitable Membranes, SinauerAssociates, Sunderland, MA, 2001. With permission.)



2.8 The family of voltage gated Ca channelsAlthough not involved directly in action potential generation or propaga-tion in axons, Ca channels are critically important players in all cellsincluding nerve cells. Some Ca channels help generate bursts of actionpotentials, some enable Ca2+ influx that then suppresses action potentialfiring for relatively long periods, and some are the coupling elements thattransduce presynaptic membrane potential into intracellular Ca2+ signalsthat control neurotransmitter release. The roles of Ca channels are diverseand central in cell physiology.

Calcium channels have been divided and subdivided on biophysical,pharmacological, and, most recently, molecular bases. Low-voltage-activated(LVA) channels have been termed T-type, as they are transient and have tiny,single-channel conductances. Three have been cloned from humans and aredesignated CaV3.1, 3.2, and 3.3 (Figure 2.12). These channels activate andthen inactivate, much like the Na channels.

High-voltage-activated (HVA) channels include all of the L-type Cachannels (CaV1.1, 1.2, 1.3, and 1.4) which are found in muscle (human geneCACNA1S), in heart and brain (CACNA1C and CACNA1D), and at synapticterminals of photoreceptors and bipolar cells in the retina (CACNA1F). TheCaV2.1, 2.2, and 2.3 group are responsible for synaptic transmission through-out the nervous system. Generally, HVA Ca channels show little inactivation.

Figure 2.12 Classification of Ca channel proteins and human genes. The low-volt-age-activated, T-type Ca channels are very similar kinetically to Na channels show-ing rapid inactivation and are involved in generating bursts of action potentials.The high-voltage-activated channels, further subdivided into L-type and P/Q-, N-,and R-type, have very slow inactivation or none at all. L-type channels are foundin muscle (CACNA1S), as well as photoreceptors in the retina (CACNA1F), and inmany cases allow the influx of Ca2+ that leads to important cellular responses suchas the release of neurotransmitter. (From Ertel, E. et al., Neuron, 25, 533–535, 2000.With permission.)



Calcium channel structure is nearly the same as Na channel structure(Figure 2.7). As is the case for the Na channels, the four hydrophobic loopslinking TM5 and TM6 of each repeating motif form the pore of the Cachannels. In the case of the L-type Ca channel, four glutamate residues, onefrom each motif, form a negatively charged ring that serves as the high-affinity binding site for Ca2+ ions.7 Mutation of these glutamate residuesreduced the selectivity of the channel for Ca2+, in effect turning it into a Na+

channel. Indeed, in the absence of divalent cations, most Ca channels conductmonovalent cations such as Na+ or Li+ very effectively. Thus, the ring ofglutamates is essential for Ca2+ binding and it is this property that definesthe Ca channels.

2.9 ConclusionWhile the voltage-gated ion channels form families and the strong evolu-tionary links can be seen between these families, an enormous degree ofdiversity is still reflected in the structure of channels. The contribution tomembrane excitability from other channel types should be considered.There are several classes of voltage-gated Cl channels that neurons expressas well as voltage-gated nonselective cation channels. These are generallypermeable to Na+ and K+ but retain some Ca2+ permeability as well, whichmay be important in intracellular Ca2+ signaling. In photoreceptors, whichare considered to be nonspiking cells, nonselective cation channels acti-vated by hyperpolarization play a prominent role in shaping the lightresponse. Generally, these types of channels are not highly expressed inaxons; therefore, their role in the transduction of neuroprosthetic signalsmay be extremely limited.

References1. Ashcroft, F.M., Ion Channels and Disease, Academic Press, San Diego, CA, 2000.2. Armstrong, C.M., Inactivation of the potassium conductance and related phe-

nomena caused by quaternary ammonium ion injected in squid axons, J. Gen.Physiol., 58, 553–575, 1969.

3. Hall, Z.W., An Introduction to Molecular Neurobiology, Sinauer Associates, Sun-derland, MA, 1992.

4. Heinemann, S.H., Terlau, H., Stuhmer, W., Imoto, K., and Numa, S., Calciumchannel characteristics conferred on the sodium channel by single mutations,Nature, 356, 441–443, 1992.

5. Hille, B., Ion Channels of Excitable Membranes, Sinauer Associates, Sunderland,MA, 2001.

6. Hodgkin, A.L. and Huxley, A.F., The components of membrane conductancein the giant axon of Loligo, J. Physiol., 116, 473–496, 1952.

7. Yang, J., Ellinor, P.T., Sather, W.A., Zhang, J.F., and Tsien, R.W., Moleculardeterminants of Ca2+ selectivity and ion permeation in L-type Ca channels,Nature, 366, 158–161, 1993.



chapter three

Neuron modeling

Frank Rattay, Robert J. Greenberg, and Susanne Resatz

Contents

3.1 Introduction3.2 Models for the cell membrane

3.2.1 Axon models of the Hodgkin–Huxley type3.2.2 Influence of temperature3.2.3 Compartment models3.2.4 Activating function3.2.5 Activating function for long fibers3.2.6 Membrane current fluctuations

3.3 The electrically stimulated retina3.4 Concluding remarksReferences

3.1 IntroductionA breakthrough in our understanding of the physics of neural signals, whichpropagate as membrane voltage along a nerve fiber (axon), was achieved bythe ingenious work of Hodgkin and Huxley1 on the nonmyelinated squidaxon. This work helped explain the action potential or “spike” that conveysthe all-or-none response of neurons. To explore the complicated gating mech-anism of the ion channels, the stimulating electrode was a long uninsulatedwire, thus every part of the neural membrane had to react in the same way;that is, propagation of signals was prevented (Figure 3.1A). Refinements oftheir method as well as the application of patch clamp techniques havesupplied us with models for different neural cell membranes. Reliable pre-diction of membrane voltage V as a function of time is possible for arbitrarystimulating currents Istimulus with proper membrane models.

1100_ch03.fm Page 39 Wednesday, October 16, 2002 8:45 AM


The main equation for internal stimulation of the soma, or any othercompartment where current flow to other processes or neighbored compart-ments is prevented, always has the same form: One part of the stimulatingcurrent is used to load the capacity Cm (in Farads) of the cell membrane andthe other part passes through the ion channels; that is,

(3.1)

The rate of membrane voltage change, dV/dt, follows as:

(3.2)

where the ion currents Iion are calculated from appropriate membrane mod-els. Usually, the membrane models are formulated for 1 cm2 of cell membraneand the currents in Eq. (3.1) become current densities.

Figure 3.1 Stimulation without signal propagation (space clamp). A stimulating elec-trode in the form of an uninsulated wire inserted along an unmyelinated nerve fiber(A) or current injection in a spherical cell (B) causes the same inside voltage, Vi, forevery part of the cell membrane. (C) In the first subthreshold phase, the voltageacross the cell membrane follows the simple constant membrane conductance model(dashed line). Figure shows simulation of a 30-µm spherical cell sheltered by amembrane with squid axon ion channel distribution (i.e., by solving theHodgkin–Huxley model with original data). Membrane voltage V is the differencebetween internal and external voltage: V = Vi – Ve; Ve = 0.

ELECTRODE

ELECTRODE

50mV

5ms

A UNMYELINATED AXON

B SPHERICAL CELL

C

Constant membrane conductance

Hodgkin-Huxleymembrane dynamics

stimulus=100pAI

Ve Vi

I CdVdt

Istimulus m ion= +

dVdt

I I Cion stimulus m= − +[ ]/

1100_Book Page 40 Tuesday, October 15, 2002 3:44 PM


A positive stimulating current applied at the inside of an axon or at anyother part of a neuron will cause an increase of V according to Eq. (3.1), ifthe membrane has been in the resting state (Istimulus = 0 and dV/dt = 0) before.In order to generate a spike, this positive stimulus current has to be strongenough for the membrane voltage to reach a threshold voltage, which causesmany of the voltage-sensitive sodium channels to open. By sodium currentinflux, the transmembrane potential increases to an action potential withoutthe need of further stimulating support. This means that as soon as the solidline in Figure 3.1C is some few millivolts above the dashed line we canswitch off the stimulus without seeing any remarkable change in the shapeof the action potential.

In general the excitation process along neural structures is more com-plicated than under space clamp conditions as shown in Figure 3.1. Currentinflux across the cell membrane in one region influences the neighboringsections and causes effects such as spike propagation (Figure 3.2). Besidesmodeling the natural signaling, the analysis of compartment models helpsto explain the influences of applied electric or magnetic fields on represen-tative target neurons. Typically, such a model neuron consists of functionalsubunits with different electrical membrane characteristics: dendrite, cellbody (soma), initial segment, myelinated nerve fiber (axon), and nonmyeli-nated terminal. Plenty of literature exists on stimulated fibers, but little hasbeen written about external stimulation of complete neurons.

In 1976, McNeal2 presented a compartment model for a myelinated nervefiber and its response to external point-source stimulation. He inspired manyauthors to expand his model for functional electrical stimulation of theperipheral nerve system, such as analysis of external fiber stimulation bythe activating function,3,4 unidirectional propagation of action potentials,5stimulation of a nerve fiber within a6,7 selective axon stimulation,8,9,10 andinfluence of fiber ending.11 Of specific interest is simulation of the thresholdand place of spike initiation generated with stimulating electrodes (e.g., bythe near field of a point source or dipole) by finite element calculation for aspecific implanted device or when the farfield influence from surface elec-trodes is approximated by a constant field.12 Stronger electric stimuli orapplication of alternating currents generate new effects in neural tissue. Alleffects depend essentially on the electric properties of the neural cell mem-brane and can be studied with compartment modeling. The ion channeldynamics can be neglected during the first response of the resting cell, butthe complicated nonlinear membrane conductance becomes dominant in thesupra-threshold phase (Figure 3.1C). Consequently, the behavior can be ana-lyzed with simple linear models or with more computational effort by sys-tems of differential equations that describe the ion channel mechanisms inevery compartment individually.

Modeling the sub-threshold neural membrane with constant conduc-tances allows the analysis of the first phase of the excitation process by theactivating function as a rough approach. Without inclusion of the compli-cated ion channel dynamics, the activating function concept explains the



basic mechanism of external stimulation, the essential differences betweenanodic and cathodic threshold values, its dependence on the geometric sit-uation, the mechanism of one side firing, and the blocking of spike propa-gation by hyperpolarized regions, as well as several other phenomena.13,14

Beside nerve fiber analysis, the activating function, which represents thedirect influence of the electric field in every compartment, is useful formagnetic stimulation15,16 and for direct stimulation of denervated musclefibers17 or, in generalized form, cardiac tissue18,19 and arbitrary neurons.20,12

In previous work we have shown how the shape of a neuron affects theexcitation characteristics and that several surprising phenomena may occur.

Figure 3.2 Simulation of the natural excitation of a human cochlear neuron by stim-ulation with a 50-pA, 250-

µsec current pulse injected at the peripheral end. Thisbipolar cell is a non-typical neuron: (1) the dendrite is myelinated and often calledperipheral axon; (2) in contrast to animal cochlear neurons, the soma and the pre-and post-somatic regions are unmyelinated; (3) a single synaptic input from theauditory receptor cell (inner hair cell) generates a spike that propagates with aremarkable delay across the current consuming somatic region. Note the decay ofthe action potential in every internode that again is amplified in the next node ofRanvier. Simulation uses internode with constant membrane conductance, a “warm”Hodgkin–Huxley model (k = 12) in the active membranes of the soma, and a 10-foldion channel density in the peripheral terminal (node 0), all nodes, and pre- and post-somatic compartments. The lines are vertically shifted according to their real locationof the rectified neuron. For details, see Rattay et al.35



This is demonstrated, for example, by comparing the threshold current forneurons with small and large soma. Whereas thick axons are easier to stimulatethan thin ones (known as the inverse recruitment order21), this relation doesnot hold for the size of the soma: a 100-

µs pulse from a point electrode 430

µm above the soma requires –2.3 mA to stimulate a 30-

µm-diameter somaneuron with a 2-

µm axon but only –1.1 mA for a 10-

µm soma. Increasing theelectrode distance to 1000

µm results in the same –2.2-mA threshold for bothcases; that is, the second surprise is that for the large soma the neuron excita-tion threshold increases slightly when the electrode is moved within a specificrange toward the soma.12 The explanation is that the axon, which is generallymore excitable than the soma, loses more current to load the larger capacityof the large soma; this effect is more pronounced when the place of spikeinitiation within the axon is close to the soma.

Comparing a pyramidal cell with a cochlear neuron demonstrates thevariety in architecture and signal processing principles in the dendrite andsoma region. Some neuron types, such as the afferent bipolar cochlear neurons,are effective transmission lines, where nearly all of the spikes initiated at thesynaptic contact with an inner hair cell arrive with some delay and smalltemporal variation (jitter) in the terminal region (Figure 3.2). In contrast to thecochlear neuron, a single synaptic input at the dendritic tree of a pyramidalcell produces only minimal change in membrane voltage at the soma and atthe initial segment, and usually the collective effect of many synaptic activitiesis necessary to influence spike initiation significantly. Typically, the pyramidalcell response is dominated by the internal calcium concentration, whichdepends on two types of voltage-dependent calcium channels: (1) low-voltage-activated channels respond in the subthreshold range and may include gen-eration of low-threshold spikes, and (2) high-voltage-activated channels in thedendrites respond, for example, to backpropagating sodium spikes.22–24

3.2 Models for the cell membraneThe cell membrane is not a perfect insulator for ion transport. When one Nachannel opens for a short time, some thousand sodium ions may be driveninto the cell by the high outside concentration. The second driving elementis the voltage across the membrane, which requires a Na+ current accordingto Ohms law. The conductance of a patch of cell membrane for a specific iontype depends on the number of open channels and is defined by Ohm’s lawand a battery voltage according to the Nernst equation. A different internalconcentration, ci, and external ion concentration, ce, cause the membranevoltage, Em, when one type of ions is involved:

(3.3)

where the gas constant R = 8.31441 J/(mol.K); temperature, T, is in Kelvin;z is valence and the Faraday constant represents the charge of one mol of

ERTzF

ccm

e

i

= ln



single-valenced ions F = 96485 c/mol. Note: At room temperature (T =20°C = 293.15 K), the factor RT/F is about 25 mV.

In their experiments Hodgkin and Huxley1 found the sodium currentproportional to E – ENa. Approximating the Nernst equation [Eq. (3.3)] withtheir data gives ENa = 45 mV = (25 mV)ln(ce/ci) and results in ce/ci =exp(45/25) = 6.05; that is, the outside sodium concentration is six times theinner one, and EK = –82 mV = (25 mV)ln(ce/ci) requires that the potassiumoutside:inside concentration is 1:26.5.

The Goldman equation defines the steady state membrane voltage Em =Erest when several ion types are involved. When K+, Na+, and Cl– ions areconsidered it reads as:

(3.4)

where [K] is the potassium concentration and the suffixes i and e stand forinside and external, respectively. PK, PNa, and PCl are permeabilities measuredin centimeters per second (cm/sec). Note that sodium and potassium areanions, but chloride is cathodic, therefore, [Cl]i appears in the numerator incontrast to the anionic concentrations. In the resting state, the membrane ismost permeable to potassium ions, and the resting membrane voltage ofabout –70 mV (i.e., the inside is more negative compared to the extracellularfluid) is close to the potassium Nernst potential.

The nonlinear conductance of the neural cell membrane depends ondifferent classes of ion channels25 (1 through 3, below) and on the activityof ion pumps (4, below):

1. Voltage-dependent gating dominates excitation and neural signalpropagation in the axon; open/close kinetics depend on the voltageacross the cell membrane.

2. Calcium-dependent gating occurs mainly at the soma and dendrites;the calcium concentration regulates the opening of the channels asopening depends on intracellular calcium ion binding.

3. Transmitter gating and second messenger gating occurs at the pre-and postsynaptic membrane.

4. Ion pumps are membrane molecules that consume energy pump ionsacross the cell membrane in order to restore the high individual ionconcentration on one side of the cell (e.g., high external sodium andhigh internal potassium concentration).

3.2.1 Axon models of the Hodgkin–Huxley type

The Hodgkin–Huxley model (HH model) was developed from a homoge-neous nonmyelinated squid axon. It includes sodium, potassium, and leak-age currents and has the form:

ERTF

P K P Na P ClP K P Na P Clm

K e Na e Cl i

K i Na i Cl e

=+ ++ +

ln[ ] [ ] [ ][ ] [ ] [ ]



(3.5)

(3.6)

(3.7)

(3.8)

(3.9)

where V is the reduced membrane voltage, resulting from internal, external,and resting potential: V = Vi – Ve – Vrest; gNa, gK, and gL are the maximumconductances for sodium, potassium, and leakage per cm2, respectively; m,h, and n are probabilities (with values between 0 and 1) that reduce themaximum conductances of sodium and potassium according to experimentalgating data; VNa, VK, and VL are the sodium, potassium, and leakage batteryvoltages, respectively, according to the Nernst equation [Eq. (3.3)]; istimulus isthe stimulus current (µA/cm2); c is the membrane capacity per cm2; α and βare voltage-dependent variables fitted from experimental data to quantify theion channel kinetics; k is a temperature coefficient that accelerates the gatingprocess for temperatures higher than the original experimental temperatureof 6.3°C; and temperature T is in °C. Although a temperature conversion ispossible, it is interesting to note that, unlike real tissue, the Hodgkin–Huxleyequations do not propagate action potentials above 31°C.14 Parameter values,units, and further expressions of the HH model are listed in Table 3.1.

In 1964, Frankenhaeuser and Huxley26 developed a model for the myeli-nated frog axon node (FH model) assuming HH-like gating mechanisms,but they derived the ion current formulation from the Nernst–Planck equa-tion and added a nonspecific current density, iP:

(3.10)

(3.11)

(3.12)

dVdt

g m h V V g n V V g V V i cNa Na K K L L stimulus= − − − − − − +[ ]3 4( ) ( ) ( ) /

dmdt

m km m m= − + +[ ]( )α β α

dhdt

h kh h h= − + +[ ]( )α β α

dndt

n kn n n= − + +[ ]( )α β α

k T= −30 1 0 63. .

dVdt

i i i i i cNa K P L stimulus= − − − − +[ ]/

i P m hEFRT

Na Na EF RTEF RTNa Na

o i=−−

22

1[ ] [ ] exp( / )

exp( / )

i P nEFRT

K K EF RTEF RTK K

o i=−−

22

1[ ] [ ] exp( / )

exp( / )



46H

andbook of Neuroprosthetic M

ethods

Table 3.1 Expressions and Constants for Axon Membrane Models

HH Model FH Model CRRSS Model SE Model SRB Model

αm

βm

αn

βn

αh

βh

αp

βp

Vrest (mV) –70 –70 –80 –78 –84

2 5 0 12 5 0 1 1. .

exp( . . )−− −

VV

0 36 22

122

3

. ( )

exp

VV

−

− −

97 0 363

131

5 3

+

+ −

.

exp.

VV

1 87 25 41

125 41

6 06

. ( . )

exp.

.

VV

−

− −

4 6 65 6

165 6

10 3

. ( . )

exp.

.

VV−

− − +

418

.exp −

V0 4 13

113

20

. ( )

exp

−

− −

VV

α m

Vexp

..

−

23 84 17

3 97 21

121

9 41

. ( )

exp.

−

− −

VV

0 33 61 3

161 3

9 16

. ( . )

exp.

.

−

− −

VV

0 1 0 011 0 1 1. .

exp( . )−− −

VV

0 02 35

135

10

. ( )

exp

VV

−

− −

0 13 35

135

10

. ( )

exp

VV

−

− −

0 0517 9 2

19 2

1 1

. ( . )

exp.

.

VV

+

− − −

0 12580

. .exp −

V0 05 10

110

10

. ( )

exp

−

− −

VV

0 32 10

110

10

. ( )

exp

−

− −

VV

0 092 8

18

10 5

. ( )

exp.

−

− −

VV

0 0720

. .exp −

V − +

− +

0 1 10

110

6

. ( )

exp

VV

βh

Vexp

.−

5 55

− +

− +

0 55 27 74

127 74

9 06

. ( . )

exp.

.

VV

− +

− +

0 21 27

127

11

. ( )

exp

VV

13 0 1 1exp( . )− +V

4 5

145

10

.

exp+ −

V15 6

124

10

.

exp+ −

V22 6

15612 5

.

exp.

+ −

V14 1

155 2

13 4

.

exp.

.+ −

V

0 006 40

140

10

. ( )

exp

VV

−

− −

0 0079 71 5

171 5

23 6

. ( . )

exp.

.

VV

−

− −

− +

− +

0 09 25

125

20

. ( )

exp

VV

− −

− −

0 00478 3 9

13 9

21 8

. ( . )

exp.

.

VV

1100_Book Page 46 T

uesday, October 15, 2002 3:44 PM

©2003 C

RC

Press LL

C


VNa (mV) 115 — 115 — —VK (mV) –12 — — — 0VL (mV) 10.6 0.026 –0.01 0 0gNa (kΩ–1/cm–2) 120 — 1445 — —gK,fast (kΩ–1/cm–2) 36 — — — 30gK,slow (kΩ–1/cm–2) — — — — 60gL (kΩ–1/cm–2) 0.3 30.3 128 86 60c (kΩ–1/cm–2) 1 2 2.5 2.8 2.8V(0) 0 0 0 0 0m(0) 0.05 0.0005 0.003 0.0077 0.0382n(0) 0.32 0.0268 — 0.0267 0.2563h(0) 0.6 0.8249 0.75 0.76 0.6986p(0) — 0.0049 — — 0.0049T0 6.3°C 293.15 K = 20°C 37°C 310.15 K = 37°C 310.15 K = 37°CQ10(αm) 3 1.8 3 2.2 2.2Q10(βm) 3 1.7 3 2.2 2.2Q10(αn) 3 3.2 3 3 3Q10(βn) 3 2.8 3 3 3Q10(αh) 3 2.8 3 2.9 2.9Q10(βh) 3 2.9 3 2.9 2.9PNa (cm/s) — 0.008 — 0.00328 0.00704PK (cm/s) — 0.0012 — 0.000134 —PP (cm/s) — 0.00054 — — —[Na]o (mmol/l) — 114.5 — 154 154[Na]i (mmol/l) — 13.7 — 8.71 30[K]o (mmol/l) — 2.5 — 5.9 —[K]i (mmol/l) — 120 — 155 —

Note: HH model = Hodgkin–Huxley model; FH model = Frankenhaeuser–Huxley; CRRSS model = Chiu, Ritchie, Rogert, Stagg, and Sweeney model;SE model = Schwarz–Eikhof model; and SRB model = Schwarz, Reid, and Bostock model. Faraday constant F = 96485 C/mol; gas constant R =8314.4 mJ/(mol.K).

1100_Book Page 47 T



(3.13)

(3.14)

(3.15)

(3.16)

(3.17)

(3.18)

(3.19)

Note that membrane voltage is denoted as E in Eqs. (3.11) to (3.13) andas V (reduced membrane voltage) in Eqs. (3.10) and (3.14). Temperature Tis measured in degrees K; for temperatures other than 20°C = 293.15 K, theα and β values in Eqs. (3.16) to (3.19) must be modified. Parameter valuesand further expressions of the FH model are listed in Table 3.1. Sodiumcurrent plays the dominant role in the action potential of the mammaliannode of Ranvier, and in contrast to the axon model of squid and frog thereare almost no potassium currents.27–29

The CRRSS model, named after Chiu, Ritchie, Rogert, Stagg, andSweeney, describes a myelinated rabbit nerve node extrapolated from orginal14°C data to 37°C:28,30

(3.20)

(3.21)

(3.22)

(3.23)

i P pEFRT

Na Na EF RTEF RTP P

o i=−−

22

1[ ] [ ] exp( / )

exp( / )

i g V VL L L= −( )

E V Vrest= +

dmdt

mm m m= − + +( )α β α

dndt

nn n n= − + +( )α β α

dhdt

hh h h= − + +( )α β α

dpdt

pp p p= − + +( )α β α

dVdt

g m h V V g V V i cNa Na L L stimulus= − − − − +[ ]2 ( ) ( ) /

dmdt

m km m m= − + +[ ]( )α β α

dhdt

h kh h h= − + +[ ]( )α β α

k T= −30 1 3 7. .



Note that the temperature factor k = 1 for T = 37°C. Parameter values andfurther expressions of the CRRSS model are listed in Table 3.1.

Schwarz and Eikhof obtained a model of FH type from voltage clampexperiments on rat nodes.29 From the original data, the Schwarz–Eikhof (SE)model results by assuming a nodal area of 50 µm2.14,31

(3.24)

(3.25)

(3.26)

(3.27)

(3.28)

(3.29)

(3.30)

(3.31)

Parameter values and additional expressions of the SE model are listed inTable 3.1.

Schwarz et al.33 derived the SRB model from human nerve fibers at roomtemperature. Single-action potentials are little effected by removing the fastor slow potassium currents, but a slow K conductance was required to limitthe repetitive response of the model to prolonged stimulating currents. SRBmodel follows:

(3.32)

(3.33)

dVdt

i i i i cNa K L stimulus= − − − +[ ]/

i P m hEFRT


o i=−−

32

1[ ] [ ] exp( / )

exp( / )

i P nEFRT

K K EF RTEF RTK K

o i=−−

22

1[ ] [ ] exp( / )

exp( / )

i g V VL L L= −( )

E V Vrest= +

dmdt

mm m m= − + +( )α β α

dndt

nn n n= − + +( )α β α

dhdt

hh h h= − + +( )α β α

dVdt

i i i i i cNa K fast K slow L stimulus= − − − − +[ ]/, ,

i P m hEFRT


o i=−−

32

1[ ] [ ] exp( / )

exp( / )



(3.34)

(3.35)

(3.36)

(3.37)

(3.38)

(3.39)

(3.40)

(3.41)

Additional SRB model data for 37°C are listed in Table 3.1. (Axon mem-brane models for biomedical applications are further discussed in References14 and 31 through 36.)

Some authors neglect the weak K currents in the mammalian axontotally because of their small amounts, but up to five types of K ionchannels are shown to have notable influences on axonal signaling, espe-cially in the internode, an element often modeled as perfect insulator withcapacity C = 0. Curious effects such as spontaneous switching betweenhigh and low threshold states in motor axons are suggested to be aconsequence of K+ loading within small internodal spaces.37 Phenomenasuch as different excitability fluctuations after action potentials in sensoryand motor neurons require far more sophisticated modeling. This meansthat for many applications the modeler’s work is not finished by selectingthe SRB model (as it is based on human data) or by fitting its parametersaccording to the shape and propagation velocity of a measured actionpotential.

The dendrite and especially the soma region is more difficult to simulatebecause a variety of ion channel types are involved.37 Some membrane mod-els are available,39–45 but even in these models we cannot rely on a preciseindividual measurement of all the ion components that vary in channeldensity and open/close kinetics. Repetitive firing (bursting) may occur asresponse to stimulation or in pacemaker neurons without any input.

i g n V VK fast K K, ( )= −4

i g p V VK slow K slow K, , ( )= −

i g V VL L L= −( )

E V Vrest= +

dmdt

mm m m= − + +( )α β α

dndt

nn n n= − + +( )α β α

dhdt

hh h h= − + +( )α β α

dpdt

pp p p= − + +( )α β α



3.2.2 Influence of temperature

Most of the membrane model data were gathered at low temperatures, butfor biomedical applications they have to be extrapolated to physiologicaltemperatures. Usually, a specific constant, Q10, is introduced to account forthe acceleration in membrane-gating dynamics when the temperature isincreased by 10°C. Hodgkin and Huxley used Q10 = 3 as a common coeffi-cient for all gating variables m, n, and h [Eqs. (3.6)–(3.8)]:

(3.42)

Raising the temperature causes shortening of the action potential and anincrease of the spike propagation velocity in the unmyelinated squid axon.The spike duration in Figure 3.1 with original HH data (6.3°C) is about tentimes longer than that of the cochlea neuron in Figure 3.2, which also wassimulated with HH kinetics. The temporal membrane behavior in cat46 wasfitted with k = 12, corresponding to a temperature of 29°C (Eq. (3.42)); seeMotz and Rattay47 and Rattay,14 p. 165). The HH spike amplitude becomesrather small for higher temperatures. An essentially reduced amplitude isnot able to preserve the excitation process along the fiber, and for tempera-tures higher than 31 to 33°C action potentials will not propagate any morein squid axons (heat block).14,48,49 Such amplitude reductions are not seen inthe membrane models of myelinated axons (FH, CRRSS, SE, and SRB mod-els), and the heat block phenomenon is absent both in myelinated andunmyelinated fibers of warm-blooded animals. In all axon models, spikeduration shortens considerably when temperature is increased — for exam-ple, from 20°C (usual for data acquisition) to 37°C.14,32 A temperature factor,k = 12, in the HH-model fits the temporal characteristics of action potentialsin unmyelinated axons of humans and warm-blooded animals at 37°C.14,35,50

The original FH model51 incorporates thermal molecular motion accord-ing to the laws of gas dynamics [Eqs. (3.11)–(3.13)], but the authors did notinclude the necessary Q10 values presented in Table 3.1 for the gating variablesm, n, h, and p. The gating process becomes essentially accelerated for hightemperatures and dominates spike duration.14,32,48,52 Also, the threshold cur-rents are influenced by the temperature: The warmer membrane is easier toexcite (Figure 3.3). Therefore, the Q10 factors have to be involved in functionalelectrical nerve stimulation modeling. Note that many published resultsobtained with the original FH model must be corrected to be valid for 37°C.

3.2.3 Compartment models

Small pieces of a neuron can be treated as isopotential elements and a wholeneuron is represented by an electric network (Figure 3.4). A current injectedat the nth compartment has to cross the membrane as capacitive or ioniccurrent and leave to the left or right side as longitudinal current; that is,application of Kirchhoff’s law is an extension of Eq. (3.1):

k Q T T= −10

100( )/



Figure 3.3 Threshold currents as a function of pulse duration (strength durationrelation). A myelinated fiber of diameter d and internodal length ∆x = 100d is stim-ulated by a point source located at a distance ∆x above a node. Excitation is easierfor high temperatures (full lines) and cathodic pulse. The threshold current for longpulses becomes time-independent and is called rheobase. The pulse duration belong-ing to the doubled rheobase current is called chronaxie. Simulations use the FH model;internode membrane is assumed to be a perfect insulator, with c = 0.

Figure 3.4 Scheme of a neuron with subunits and part of the simplified equivalentelectrical network (batteries resulting from different ion concentrations on both sidesof the membrane are not shown).

37°C

20°C

20°C

RHEOBASE

2*RHEOBASE

CHRONAXIE

37°C

1000

100

100.01 0.1 1msPULSE DURATION

/I delectrode[µA/µm]

Ve,2

Vi,2Vi,0

Ve,0

R/2R/2

GmCm

Vi,1

Ve,1OUTSIDE

MEMBRANE

INSIDE

DENDRITE

SOMAINITIAL SEGMENT

NODE UNMYELINATEDINTERNODE AXON



(3.43)

where n indicates the nth compartment, C is its membrane capacity, and R/2is the internal resistance between the center and the border of the compart-ment. Introducing the reduced membrane voltage V = Vi – Ve – Vrest leadsto the following system of differential equations for calculating the timecourses of Vn in every compartment:

(3.44)

The dots in Eq. (3.44) stand for terms that have to be added in cases ofmore than two neighbor elements (e.g., at the cell body [soma] or at branch-ing regions). For the first and last compartments, Eq. (3.44) has a reducedform. The membrane surface, An, of every compartment has to be calculatedto find Cn = Ancn (cn is the specific membrane capacitance) and Iion = Aniion.The ionic membrane current density, iion, is computed with an appropriatemembrane model; for cylinder elements (d, diameter; ∆x, length), we obtainAn = dnπ∆xn, , where the internal resistivity ρi is oftenassumed as ρi = 0.1 kΩcm.

In a spherical cell body with several processes (as in Figure 3.4), theinternal resistances to the neighbor compartments depend on the compart-ment diameters; that is, Rsoma→dendrite1/2 < Rsoma→axon/2 if ddendrite1 > daxon. In moredetail, Asoma = 4rsoma

2π − Σ (2rsomaπhj), with subscript j indicating the jth process)such that hj = rsoma – zj, where . The somatic resis-tance to the border of the jth process is:

(3.45)

The electric and geometric properties change from compartment to compart-ment and the modeler has to decide about the degree of complexity thatshould be involved. For example, the surface area of a dendrite compartmentand the soma can be simulated by a cylinder and a sphere or an additionalfactor is included to represent the enlargement by spines or windings. Halterand Clark53 and Ritchie54 presented detailed ion current data for the intern-ode, whereas many authors neglect internodal membrane currents. A com-promise is to simulate the internode as a single compartment with bothmembrane conductance Gm and membrane capacity C proportional to 1/N,where N is the number of myelin layers.12

I Cd V V

dtI

V V

R R

V V

R Rinjected n ni n e n

ion ni n i n

n n

i n i n

n n,

, ,,

, , , ,( )

/ / / /=

−+ +

−+

+−+

−

−

+

+

1

1

1

12 2 2 2

dVdt

IV V

R RV V

R R

V VR R

V VR R

I C

nion n

n n

n n

n n

n n

e n e n

n n

e n e n

n ninjected n n

= − + −+

+ −+

+

−+

+ −+

+ +

−

−

+

+

−

−

+

+

[

+

,

, , , ,,

/ / / /...

/ / / /... ]/

1

1

1

1

1

1

1

1

2 2 2 2

2 2 2 2

R x di n n/ /( )2 2 2= ρ π∆

z r dj soma process j= − ( )2 22, /

Rr

r zr z

soma j i soma j

soma j

,ln

2 2= +

−

ρπ



Synaptic activation of a neuron can be simulated by current injection atthe soma or at dendrite compartments (Figure 3.2). In such cases, all externalpotentials, Ve, are assumed to be 0 in Eq. (3.44). On the other hand, neuro-prosthetic devices generate neural activities by application of electric fields.Instead of current injection, the terms including the external potentials Ve

become the stimulating elements in Eq. (3.44).

3.2.4 Activating function

The driving term of the external potential on compartment n (Eq. (3.44)) iscalled the activating function fn:

(3.46)

The physical dimension of fn is V/sec or mV/msec. If the neuron is in theresting state before a stimulating current impulse is applied, the activatingfunction represents the rate of membrane voltage change in every compart-ment that is activated by the extracellular field. That is, fn is the slope ofmembrane voltage Vn at the beginning of the stimulus. Regions with positiveactivating function are candidates for spike initiation, whereas negative acti-vating function values cause hyperpolarization.

The temporal pattern generated in a target neuron by a single activeelectrode of a cochlear implant is analyzed in the following with the acti-vating function concept. The electrode is assumed as an ideal point sourcein an infinite homogeneous extracellular medium, which results in thesimple relation:

(3.47)

where Ve is measured at a point with distance r from the point source andρe is the extracellular resistivity.

The insert of Figure 3.5A shows the values of Ve and f along a cochlearneuron when stimulated with 1 mA from a monopolar electrode. The alter-nating sequence of positive and negative activating function values indicatesfirst spike initiation at peripheral node P2 (Figure 3.5B) and a second spikeinitiation region in the central axon for stronger positive monophasic stimuli(Figure 3.5C). Cathodic stimulation also generates two regions of spike ini-tiation — first, excitation in P1 (Figure 3.5D), then excitation in P4 (Figure3.5E). The activating function predicts easier excitation with cathodic cur-rents with values of –3890 V/sec vs. 2860 V/sec in P1 and P2, respectively.Note, however, that f quantifies the very first response, which is flattenedby the capacity (compare dashed line in Figure 3.1) and is further influencedby the neighbor elements.

fV V

R RV V

R RCn

e n e n

n n

e n e n

n nn= −

++ −

++−

−

+

+[ , , , ,

/ / / /...]/1

1

1

12 2 2 2

VI

ree electrode=

ρπ

.4



3.2.5 Activating function for long fibers

In 1976, McNeal introduced the first efficient model for computing peripheralnerve fiber responses in functional electrical nerve stimulation. Manyauthors followed his simplification of an ideal internode membrane, whereboth membrane conductance and membrane capacity are assumed to be 0.In this case, the main equation, Eq. (3.44), reduces to:4,12

(3.48)

Figure 3.5 Human auditory nerve response evoked by a stimulating impulse from acochlear implant. (A) Position of a cochlear neuron relative to a spherical electrode(same geometry as in Figure 3.2; unmyelinated terminal P0, nodes P1–P5 of periph-eral myelinated thin process, unmyelinated region around the soma, and a thickercentral axon). Extracellular resistivity ρe = 0.3 kΩcm, and calculation of potential froma 1-mA point source [Eq. (3.47)] results in 1 V at r = 0.24 mm, which is the electroderadius. The values of the activating function f shown in insert (A) are proportionalto the slopes of the neural response curves at stimulus onset in (B) and (C). Thelargest f value of P2 predicts the place of spike initiation in (B). All slopes of membranevoltage in the central nodes are positive and strengthen the stimulus current to 1.2mA, causing additional spike initiation at peripheral node C2 (C). The activatingfunction value of P0 is also positive, but the much stronger negative activity in P1compensates the excitation process quickly; insert (C) shows the part of the mem-brane voltage of the P0 compartment during the stimulus pulse as marked by arrows,with 5× magnification. In part (D), –800 µA per 100-µsec pulse initiates a spike at P1which is in accordance with the activating function concept. In part (E), the smallerintensity of f at P4 allows spike initiation at –1200 µA. The P4 spike develops afterthe P1, but nevertheless the P4 spike activates the central axon. Note the differentarrival times at the measuring electrode C5.

dVdt

id x

LV V V

xV V V

xc

nion n

i

n n n e n e n e n= − + − + + − +

− + − +[ ,, , , ]/∆

∆ ∆42 21 1

2

1 1

2ρ



with constant fiber diameter d, node-to-node distance ∆x, node length L, andaxoplasma resistivity ρi; both membrane capacity c and nodal ion current iion

are per cm2. The activating function:

(3.49)

becomes proportional to the second difference of the extracellular potentialalong the fiber.

Equations (3.48) and (3.49) model the unmyelinated fiber when thelength of the active membrane is equal to the compartment length: L = ∆x.With ∆x → 0, Eq. (3.49) reads as:

(3.50)

Equations (3.49) and (3.50) verify the inverse recruitment order; that is,thick fibers are easier to stimulate3,21 (Figure 3.5) by the linear relationbetween d and f.

For a monopolar point source in an infinite homogeneous medium, theactivating function is easy to evaluate. For example, for a straight axon (Eq.(3.47); Figure 3.6A,B), we obtain:

(3.51)

where el stands for electrode and z measures the center-to-center distancebetween a small spherical electrode and the axon, and (see Eq. (3.50)):

(3.52)

Equation (3.50) implies the relation between the curvature of the Ve graphand the regions of excitation in a target fiber; solving f(x) = 0 gives the pointsof contraflexure of Ve that separate the regions of depolarization (f > 0) fromthe hyperpolarized ones (f < 0). In our example, this results in the verticaldashed lines at x = xel ± 0.707zel. In the excited (f > 0) regions, the Ve graphis always above its tangents (Figure 3.6B; regions I and III for anodic stim-ulation). Strong curvature of Ve means large f values; therefore, the centralpart of region II becomes extremely hyperpolarized at anodic stimulation(Figure 3.6C). In relation to the shape of the activating function, the thresholdfor cathodic stimulation is about four times weaker (Figure 3.6E and F) whichis in accordance with experimental findings.55

fd x

LcV V V

xni

e n e n e n= − +− +∆∆4

21 1

2ρ, , ,

fd

cVxi

e=∂∂4

2

2ρ

VI

x x zee el

el el= −( ) +( )−ρπ. .

42 2

0 5

fd I

cx x z x x ze el

iel el el el= −( ) +( ) −( ) −( )−ρ

ρ π. .

162

2 22 5 2 2



Deviations between the thick lines for f and for Ve at the end of the 100-µsec pulse (just above threshold) are caused by (1) an inner-axonal currentcomponent resulting from membrane currents at the other compartments,(2) exponential loading process of the membrane capacity (in the activatingfunction concept, the dashed line in Figure 3.1 is approximated by the tan-gent at stimulus onset), and (3) the beginning of the voltage-sensitive ionchannel-gating processes. Compared to the prediction by f, there is a smallenlargement of the hyperpolarized region II seen at the zero-crossings of thethick line in Figure 3.6C. This is mostly caused by the inner-axonal current,where the strong hyperpolarized central part has a negative influence at the

Figure 3.6 Stimulation of an unmyelinated axon with a monopolar electrode. Duringthe 100-µsec stimulus pulse, the activating function defines three regions of responses,separated by vertical dashed lines. (B)–(D) Vertically shifted snapshots of membranevoltages in 100-µsec steps. Thick lines demonstrate the similarity between the acti-vating function and the membrane voltage at the end of the 100-µsec stimulus signal.Simulation is of HH data at 29°C, ρe = 0.3 kΩ, electrode fiber distance z = 1 mm, andfiber diameter 10 µm.

ELECTRODE

AXIS OF AXON

70∞

V =e

Ve

I = 6mAelectrode

I = 1.5mAelectrode

I = -1.5mAelectrode

f

f

A

B

C

D

E

F

REGION I II III

XZ

r Ie electrode24p (x-x ) +zelectrode electrode

2

r = 2 (x-x ) +zelectrode electrode2

500mV

1mm



borderlines to regions I and III. For the same reason the points of spikeinitiation (Figure 3.6C) are lateral to the maximum points of f (Figure 3.6B).

With some computational effort, the linear theory can be improved byconsidering the influences of the inner-axonal current and the exponentialloading process of the membrane capacity.56 Especially for close electrodefiber distances, the subthreshold steady-state membrane voltage will con-siderably deviate from the first response.57,58 But, the activating functionconcept demonstrates the instructive relation between a simple formula andobserved phenomena. For example, Eq. (3.49) indicates that a straight partof an axon in a constant field region is not stimulated directly; however, thebending part in a constant-field region will be de- or hyperpolarized,depending on the orientation of the axon and field. Such effects are studiedin spinal root stimulation (see Figure 3.3 of Reference 59).

Note that the equations of this section require constant data along thefiber. At a fiber ending, widening or branching the treatment is accordingto the description of Eq. (3.44). The fiber-end influence may be modeledby a boundary field driving term in addition to the activating function.11

In some applications (heart and skeletal muscle, nerve bundle), manyexcitable fibers are closely packed and the effects of current redistributionare handled by the bidomain model. The tissue is not assumed to be adiscrete structure but rather two coupled, continuous domains: one for theintracellular space and the other for the interstitial space. This space-averaged model is described with a pair of coupled partial differentialequations that have to be solved simultaneously for the intracellular andinterstitial potentials.7,18,19,60 Despite the complex situation, several stimu-lation phenomena of the heart muscle are already predicted by the activat-ing function concept from single-fiber analysis.60

3.2.6 Membrane current fluctuations

Single-fiber recordings show variations in the firing delays even when stim-ulated with current pulses of constant intensity.62,63 The main component ofthe stochastic arrival times (jitter) is explained by ion channel current fluc-tuations. Models of the Hodgkin–Huxley type define the gating variables m,n, and h as probabilities, but no variance is included. The addition of a simplenoise term in single compartment modeling can fit several of measuredcochlear nerve data.47 Other phenomena — for example, a bimodal distri-bution of spike latencies, as discussed later (Figure 3.7B) — require fullcompartment analysis with a stochastic term in every active compartment.35

Including membrane current fluctuations is significant in the followingapplications: (1) modeling temporal firing patterns (e.g., in the cochlearnerve), (2) detailed analysis of fiber recruitment (for example, constant stim-ulation conditions generate three types of axons within a nerve bundle: onegroup fires at every stimulus, the second fires stochastically, and the thirdshows solely subthreshold responses), (3) the transition from the myelinatedto the unmyelinated branching axon is less secure in regard to signal transfer,



and spikes can stochastically be lost50,64, (4) during the integration of neuralinputs of an integrate-and-fire neuron.

The elaborate models of single-channel current influences are recom-mendable for detailed investigations.65–67 Combining this work with a com-partment model will show most of the noisy behavior in nerve fibers as

Figure 3.7 Simulation of ion current fluctuations in the cochlear neuron membrane.(A) Membrane voltage in the resting (left) and activated (right) neuron in a humanand cat. Stimulation by current injection at the lateral end is close to natural spikeinitiation by synaptic activation. As a consequence of the RC circuit properties ofthe network, the voltage fluctuations become low pass filtered and neighboringcompartment behavior is correlated. The unmyelinated terminals are assumed tohave the same 1-µm diameter with same ion channel density as the peripheralnodes, but a length relation of 10 to 1.5 µm results in 2.58 times stronger membranecurrent fluctuation at the endings. The strong influence of the first compartmentto the network is still seen in the cat soma region (left) but it is lost in the humancase because of the longer peripheral axon (five nodes and a large unmyelinatedsoma region). Voltage fluctuations in the soma region are generally smaller thanin the nodes. Loading the large capacity of the unmyelinated human soma requiresthe spike about 400 µsec to cross this region, which is three times the value of themyelinated cat case (right). (B) Responses of the fifth node in the central humanaxon (compartment 25 marked as C5 in Figure 3.5) to anodic external stimulationwith 100 µsec pulses (10 runs for every stimulus intensity; situation shown in Figure3.5). Weak stimuli fail sometimes (600 µA: 8 of 10) and have larger jitter, and theirarrival is more delayed, as they are generated in the periphery. At 900 µA, spikesare also initiated in the central axon. Note that this experimentally known bimodalresponse is a consequence of ion channel current fluctuations. For details, see Rattayet al.35 and Rattay.70



reported by Verween and Derksen.66 Membrane noise increases with devia-tion from the resting voltage and is greater for hyperpolarization than fordepolarization. To reduce the enormous computational effort of simulatingevery channel activity we follow a suggestion of Rubinstein69 and assumethe noise current to be proportional to the square root of the number ofsodium channels.35,70 At every compartment, a Gaussian noise variable,GAUSS (mean = 0, standard deviation = 1), is introduced that changes itsvalue every 2.5 µsec. The amplitude of the noise term is proportional to thesquare root of the number of sodium channels involved. Noisy membranecurrents (in µA) are defined as:

(3.53)

where An denotes the membrane area of compartment n (in cm2), and gNa

(mS cm–2) is the maximum sodium conductance. A standard factor knoise =0.05 µA.mS–1/2 common to all compartments fits the observations of Verveenand Derksen.66

Simulated voltage fluctuations are compared for cochlear neurons in catsand humans. Both neurons are assumed to have the same peripheral (1 µm)and central (2 µm) axon diameters, but in the cat the peripheral axon isshorter (distance from soma to peripheral terminal in cats is 0.3 mm; inhumans, 2.3 mm), and the cat somatic region is insulated by myelin sheets.Only the 10-µm-long peripheral terminal and the nodes are without myelin;additionally, in humans, the soma region is unmyelinated (Figure 3.2). Thefollowing effects are seen in Figure 3.7A:

1. Thick fibers show less membrane fluctuations. The peripheral com-partments (1 and 5 in cats, 1 and 11 in humans) have smaller noiseamplitudes than the central axon (11 in cats, 19 in humans). Thephenomenon observed by Verveen and Derksen66 can be explainedin a simple model with corresponding states of homogeneous my-elinated axons. During the excitation process, the nodal currentsare proportional to axon diameter, but the noisy current increaseswith the square root of diameter only (nodal length is independentof diameter).

2. Membrane voltage fluctuations in one compartment have two com-ponents: the own membrane current and the noisy inner currentsfrom the neighboring compartments. Note the correlated sinusoidal-like shapes of the three lower left cat curves. The cat myelinated somais assumed to have no sodium channels; therefore, the displayedmembrane voltage has lost the first component and becomes theaveraged neighboring compartment noise filtered by the soma ca-pacity effect.

And, as shown in Figure 3.7B:

I GAUSS k A gnoise n noise n Na, = ⋅ ⋅



3. Threshold voltage has to be redefined (e.g., as 50% probability ofgenerating a spike); effectiveness increases from 600 µA (20%) to 700µA (100%).

4. Weak stimuli have larger jitter because of the longer noise influenceduring the subthreshold part of excitation. Jitter of peripherallyevoked spikes decreases from 600 to 900 µA; the same for the centralones — from 900 to 1200 µA.

5. Spike initiation regions can be sensitive to stochastic components(900 µA) .

3.3 The electrically stimulated retinaPhotoreceptor loss due to retinal degenerative diseases such as age-relatedmacular degeneration and retinitis pigmentosa is a leading cause of blind-ness. Although these patients are blind, they possess functioning bipolar andganglion cells that relay retinal input to the brain. Retina implants are indevelopment in order to evoke optical sensations in these patients. Thesuccess of such an approach to provide useful vision depends on elucidatingthe neuronal target of stimulation. A compartmental model for extracellularelectrical stimulation of the retinal ganglion cell (RGC) has been developed.71

In this model, a RGC is stimulated by extracellular electrical fields with activechannels and realistic cell morphology derived directly from neuronal trac-ing of an amphibian retina cell.

The goal of the simulation was to clarify which part of the target cell ismost excitable. It was previously shown that patients blind from retinitispigmentosa resolve focal phosphenes when the ganglion cell side of theretinal surface is stimulated. None of the patients reported the perception ofwedges; all reported spots of light or spots surrounded by dark rings. Whereis the initiation point of the signal that causes these optical sensations? Isthe threshold to propagate an action potential down the axon lowest for anelectrode over the axon, the soma, or somewhere over the dendritic arbor,which may spread up to 500 µm in diameter and overlap the dendritic fieldof other ganglion cells?72

To explore the relative thresholds of ganglion cell axons, somas, anddendrites a mudpuppy (Necturus maculosus) RGC with a large dendritic fieldand a long axon was chosen. This cell has been mapped in three dimensions.73

The entire traced cell, except the soma, is divided into cylindrical compart-ments. The soma can be modeled as a compartmentalized sphere71 or as asingle spherical compartment as described above. The number of compart-ments was increased until the thresholds did not change by more than 1%.Finally, more than 9000 compartments ensure fine numerical solutions forthe RGC, as shown in Figure 3.8.

An extracellular field from an ideal monopolar point or disk electrodein a homogeneous medium were applied to these compartments. An elec-trical network as shown in Figure 3.4 represents the whole retinal ganglioncell. Several constants were specified based on whole-cell recording data



at 22°C, including membrane capacitance (c = 1 µF/cm2) and cytoplasmicresistance (0.11 kΩcm).75 These values are assumed to be uniform through-out the cell.

For comparison, the electric conductance of the cell membrane wasevaluated with a linear passive model, a Hodgkin–Huxley model with pas-sive dendrites (HH),1 and a model composed of all active compartmentswith five nonlinear ion channels (FCM).76 At the start of all simulations, themembrane potential is initialized to a resting potential of –70 mV. The linearpassive model reduces each patch of membrane to a simple parallel RCcircuit with a leak. The leak conductance was modeled as a battery at –70mV in series with a conductance of 0.02 mS/cm2.

Figure 3.8 A topographical view of the target cell shown with its axon projectingupwards. The soma of the retinal ganglion cell is modeled as a 24-µm sphere. Thissize was chosen to approximate the diameter of the actual traced mudpuppy soma.The electrode is modeled as an ideal point source in an infinite homogeneous medium.For calculation of Ve, see Eq. (3.47); ρe = 1/60 Ωcm, the resistivity of normal (0.9%)saline, which is similar to the resistivity of the vitreous humor.76 The height of theelectrode was held fixed at 30 µm above the center of the soma, a distance comparableto that of the retinal ganglion cells to the surface of the retina in our region of interest.The electrode positions are marked by the letters A through G. The linear distancesfrom the soma (in µm) are: (A) over the axon ~503, (B) over the axon ~130, (C) oppositethe axon ~334, (D) perpendicular to the axon ~160, (E) directly above the soma, (F)perpendicular to the axon ~302, and (G) opposite the axon ~121. The vertical distance(not shown) from the electrode to the center of the nearest compartment is (in µm):(A) 33.0, (B) 30.0, (C) 34.0, (D) 30.0, (E) 30.0, (F) 29.5, and (G) 38.0.

100µm

A

B

C

D

F

E

G



The retinal ganglion cell in many species (including humans) has anonmyelinated axon inside the eye. Some simulations were performed withthe classic description for unmyelinated axons, the Hodgkin–Huxleymodel. The Hodgkin–Huxley channel kinetics are applied to the soma andthe axon, with the constants gNa = 120 mS/cm2, ENa = 50 mV, gK = 36mS/cm2, EK = –77 mV, gl = 0.3 mS/cm2, El = –54.3 mV. The dendrites arepassive in this model.

The FCM model includes five nonlinear ion channels and one linearleakage channel.44,45,76 The kinetic model was constructed on the basis ofvoltage clamp data in retinal ganglion cells of tiger salamanders and rats.77,78

The model consists of INa and ICa; the delayed rectifier, IK; IK,A; and the Ca-activated K current, IK,Ca. IK,Ca was modeled to respond to Ca influx, and avariable-rate, Ca-sequestering mechanism was implemented to remove cyto-plasmic calcium. ICa and IK,Ca do play an important role in regulating firingfrequency. The basic mathematical structure for voltage-gating is based onthe Hodgkin–Huxley model:

(3.54)

where

The gating of IK,Ca is modeled as

(3.55)

where = 0.05 mS/cm2 and the Ca2+-dissociation constant, Ca2+diss, is

10–3 mmole/l. [Ca2+]i is allowed to vary in response to ICa. The inward flowingCa2+ ions are assumed to be distributed uniformly throughout the cell; thefree [Ca2+] above a residual level — Ca2+

res = 10–4 mmole/l — are activelyremoved from the cell or otherwise sequestered with a time-constant (τ =1.5 msec). Thus, [Ca2+]i follows the equation:

(3.56)

CdEdt

g m h E E g c E E g n E E

g a h E E g E E g E E I

m Na Na Ca Ca K K

A A K K Ca K l l stimulus

= − − − − − −

− − − − − − +

3 3 4

3

( ) ( ) ( )

( ) ( ) ( ),

g mS cm E mV g mS cm g mS cm

E mV g mS cm g mS cm E to mV

Na Na Ca K

K A l l

= = = =

= − = = = − −

50 35 2 2 12

75 36 0 005 60 65

2 2 2

2 2

, , . , ,

, , . , .

g gCa Ca

Ca CaK Ca K Ca

i diss

i diss

, ,

[ ]

[ ]=

( )+ ( )

+ +

+ +

2 2 2

2 2 21

gK Ca,

d Cadt

sIF

Ca Cai Ca i res[ ] [ ] [ ]2 2 2

2

+ + +

=−

−−

ν τ



where F is the Faraday constant, s/ν is the ratio of surface to volume of thecompartment being studied, and the factor of 2 on ν is the valency. For aspherical soma with r = 12 µm, Eq. (3.56) becomes:

(3.57)

ECa is modeled as variable according to the already mentioned Nernst equa-tion:

(3.58)

where [Ca2+]e is the external calcium ion concentration = 1.8 mmole/l. Similar to the Hodgkin–Huxley model, the rate constants for m, h, c, n,

a, and hA solve the first-order kinetic equation (Table 3.2):

(3.59)

The multi-compartment model includes representations for dendritic trees,soma, axon hillock, a thin axonal segment, and an axon distal to the thinsegment. On these parts, the five ion channels were distributed with varyingdensities, simulated by varying the value of gmax (mS/cm2) for each channel(Table 3.3).44,45,76 Using the multicompartmental simulation packageNEURON79, the action potential threshold values were computed for theelectrode positions as marked in Figure 3.8.71

Table 3.2 Rate Constants for Voltage-Gated Ion Channels

α β

d Cadt

I CaiCa i

[ ]. . ([ ] . )

220 000013 0 666667 0 0001

++= − − −

ERT

FCaCaCa

e

i

=+

+2

2

2ln[ ][ ]

dxdt

xx x x= − + +( )α β α

α m E

E

e=

− +( )−− +( )

0 6 30

10 1 30

..

βmEe= − +( )20 55 18

α hEe= − +( )0 4 50 20. βh Ee

=+− +( )

6

10 1 20.

α c E

E

e=

− +( )−− +( )

0 3 13

10 1 13

..

β cEe= − +( )10 38 18

αn E

E

e=

− +( )−− +( )

0 02 40

10 1 40

..

βnEe= − +( )0 4 50 80.

α A E

E

e=

− +( )−− +( )

0 006 90

10 1 90

..

βAEe= − +( )0 1 30 10.

α hAEe= − +( )0 04 70 20. βhA Ee

=+− +( )

0 6

10 1 40

..



A practical retinal prosthesis will perhaps have disk electrodes, becauseflat electrodes are easy to produce. To replace the point electrode source bya field from an equipotential metal disk in a semi-infinite medium, theexternal potential, Ve, under the electrode is calculated by:

(3.60)

where (r, z) is the radial and axial distance from the center of the disk incylindrical coordinates z ≠ 0.80 V0 is the potential, and α is the radius of thedisk. In simulation, both 50- and 100-µm-diameter disks required the samecathodic threshold voltage. In human trials, the current required to reachthreshold also did not vary when a monopolar stimulating electrode waschanged from 50-µm-diameter to 100-µm.71,81

3.4 Concluding remarksCalcium currents and the inside calcium concentration often have significantinfluence on the temporal spiking pattern of electrically stimulated neurons,because calcium is an important intracellular signaling molecule with rapideffects on many processes. A rather simple approach demonstrates the agree-ment between the FCM model and the firing behavior of retinal ganglioncells;44,45,76 a similar model exists for pyramidal cells.42 Detailed modelingincludes the variation of calcium concentration with distance from the cellmembrane, buffer-elements and pumps.82

Selective stimulation of neural tissue is a great challenge for biomedicalengineering. A typical example in bladder control is activation of the detru-sor muscle without activation of the urethral sphincter and afferent fiberswhen stimulating sacral roots with tripolar cuff electrodes. Analysis withthe activating function is of help for finding the polarized and hyperpo-larized regions but nonlinear membrane current modeling has to beincluded to quantify cathodic and anodic block or anodic break phenomenain selected target fibers and to determine the stimulus pulse parametersfor an operating window.14,83,84

Table 3.3 FCM Model Channel Densities at Soma, Dendrite, and Axon

gmax

Soma(mS/cm2) Dendrite

Axon(0–3%)a

Axon(3–9%)a

Axon(9–100%)a

gNa 70 40 150 100 50gCa 1.5 3.6 1.5 0 0gK 18 12 18 12 15gKa 54 36 54 0 0gK,Ca 0.065 0.065 0.065 0 0

a Percent total axon length from soma; total axon length approximately 1 mm.

V r zV a

r a z r a z( , ) arcsin

( ) ( )= ⋅

− + + + +

2 202 2 2 2π



Stimulation of neurons in a network is often difficult to explore becauseof the complex neural geometry. Every type of the involved neurons hasto be analyzed in regard to the response and its sensitivity. Note thatneurotransmitter release can be initiated or influenced by the applied fieldwithout the generation of a spike.12 This may be of specific importance atthe retina where several types of neurons never generate action potentialsbut their degree of neurotransmitter release heavily depends on transmem-brane voltage.

The extracellular potential along a target cell is input data for the exci-tation process. Some case studies are based on the combination of ideal pointsources in infinite homogeneous media or idealized surface electrodes,71,80,85

but most clinical applications require more precise calculation of the field.Finite element software such as ANSYS is commonly used70,86,87 but so, too,are finite differences6 and the boundary element method.88

Simulation tools such as NEURON (http://www.neuron.yale.edu/)79

and GENESIS (http://www.genesis-sim.org/GENESIS/)89 are recommendedfor solving neural compartment models. Compartment modeling (i.e., Eq.(3.44)), together with a set of equations for the ion currents, results in asystem of ordinary differential equations that can be solved directly. Thisprocedure (methods of lines) will cause numerical problems for fine spatialsegmentation. Integration routines for stiff differential equations, implicitintegration methods, or solving with the Crank Nicholson method are ofhelp to reduce the computational effort.90

References1. Hodgkin, A.L. and Huxley, A.F., A quantitative description of membrane

current and its application to conduction and excitation in nerve, J. Physiol.,117, 500–544, 1952.

Table 3.4 Cathodic 100-µsec Current Thresholds forPoint-Source Stimulationa

ElectrodePassive Model

HHModel

FCMModel

A 0.900 1.58 1.73B 1.00 1.79 1.97C 0.954 75.1 55.2D 0.699 14.4 2.72E 1.00 1.00 1.00F 0.812 65.1 2.80G 1.05 9.58 2.63

a Normalized for position over the soma and calculated with thethree membrane models. The absolute threshold current at thesoma (electrode position E) for the passive model was –32.9 µA;for the Hodgkin–Huxley model, –43 µA; and for the FCM (fivenonlinear ion channels) model, –71 µA. The threshold responsefor the passive model is defined as a 15-mV depolarization.13



http://www.neuron.yale.edu/

http://www.genesis-sim.org/GENESIS/

2. McNeal, D.R., Analysis of a model for excitation of myelinated nerve, IEEETrans. Biomed. Eng., BME-23, 329–337, 1976.

3. Rattay, F., Analysis of models for external stimulation of axons, IEEE Trans.Biomed. Eng., 33, 974–977, 1986.

4. Rattay, F., Analysis of models for extracellular fiber stimulation, IEEE Trans.Biomed. Eng., 36, 676–682, 1989.

5. Sweeney, J.D. and Mortimer, J.T., An asymmetric two electrode cuff for gen-eration of unidirectionally propagated action potentials, IEEE Trans. Biomed.Eng., BME-33, 541–549, 1986.

6. Veltink, P.H., van Alste, J.A., and Boom, H.B.K., Simulation of intrafascicularand extraneural nerve stimulation, IEEE Trans. Biomed. Eng., BME-35, 69–75,1988.

7. Altman, K.W. and Plonsey, R., Point source nerve bundle stimulation: effectsof fiber diameter and depth on simulated excitation, IEEE Trans. Biomed. Eng.,37, 688–698, 1990.

8. Sweeney, J.D., Ksienski, D.A., and Mortimer, J.T., A nerve cuff technique forselective excitation of peripheral nerve trunk regions, IEEE Trans. Biomed.Eng., BME-37, 706–715, 1990.

9. Veraart, C., Grill, W.M., and Mortimer, J.T., Selective control of muscle acti-vation with a multipolar nerve cuff electrode, IEEE-Trans. Biomed. Eng., 40,640-653, 1990.

10. Tyler, D.J. and Durand, D.M., Intrafascicular electrical stimulation for selec-tively activating axons, IEEE Eng. Med. Biol. 13, 575-583, 1994.

11. Nagarajan, S.S., Durand, D., and Warman, E.N., Effects of induced electricfields on finite neuronal structures: a simulation study, IEEE Trans. Biomed.Eng., BME-40, 1175–1188, 1993.

12. Rattay, F., The basic mechanism for the electrical stimulation of the nervoussystem, Neuroscience, 89, 335–346, 1999.

13. Coburn, B., Neural modeling in electrical stimulation review, Crit. Rev. Biomed.Eng., 17, 133–178, 1989.

14. Rattay, F., Electrical Nerve Stimulation: Theory, Experiments and Applications,Springer-Verlag, New York, 1990.

15. Basser, P.J., Wijesinghe, R.S., and Roth, B.J., The activating function for mag-netic stimulation derived from a 3-dimensional volume conductor model,IEEE Trans. Biomed. Eng., 39, 1207–1210, 1992.

16. Garnham, C.W., Barker, A.T., and Freeston, I.L., Measurement of the activatingfunction of magnetic stimulation using combined electrical and magneticstimuli, J. Med. Eng. Technol., 19, 57–61, 1995.

17. Reichel, M., Mayr, W., and Rattay, F., Computer simulation of field distribu-tion and excitation of denervated muscle fibers caused by surface electrodes,Artif. Organs, 23, 453–456, 1999.

18. Sobie, E.A., Susil R.C., and Tung, L., A generalized activating function forpredicting virtual electrodes in cardiac tissue, Biophys. J., 73, 1410–1423, 1997.

19. Efimov, I.R., Aguel, F., Cheng, Y., Wollenzier, B., and Trayanova, N., Virtualelectrode polarization in the far field: implications for external defibrillation,Am. J. Physiol. Heart. Circ. Physiol., 279, H1055–H1070, 2000.

20. Rattay, F., Analysis of the electrical excitation of CNS neurons, IEEE Trans.Biomed. Eng., 45, 766–772, 1998.

21. Blair, E.A. and Erlanger, J., A comparison of the characteristics of axons throughtheir individual electrical responses, Am. J. Physiol., 106, 524–564, 1933.



22. Huguenard, J.R., Low-threshold calcium currrents in central nervous systemneurons, Annu. Rev. Physiol., 58, 329–348, 1996.

23. Schiller, J., Helmchen, F., and Sakmann, B., Spatial profile of dendritic calciumtransients evoked by action potentials in rat neocortical pyramidal neurones,J. Physiol., 487, 583–600, 1995.

24. Mainen, Z.F. and Sejnowski, T.J., Modeling active dendritic processes in py-ramidal neurons, in Methods in Neuronal Modeling: From Ions to Networks, 2nded., Koch, C. and Segev, I., Eds., MIT Press, Cambridge, MA, 1998, pp.171–209.

25. Destexhe, A, Mainen, Z.F., and Sejnowski, T.J., Synthesis of models for excit-able membranes, synaptic transmission, and neuromodulation using a com-mon kinetic framework, J. Comput. Neurosci., 1, 195–230, 1994.

26. Frankenhaeuser, B., Sodium permeability in toad nerve and in squid nerve,J. Physiol., 152, 159–166, 1960.

27. Horáckova, M., Nonner, W., and Stämpfli, R., Action potentials and voltageclamp currents of single rat Ranvier nodes, Proc. Int. Union Physiol. Sci., 7,198, 1968.

28. Chiu, S.Y., Ritchie, J.M., Rogart, R.B., and Stagg, D., A quantitative descriptionof membrane currents in rabbit myelinated nerve, J. Physiol., 313, 149–166, 1979.

29. Schwarz, J.R. and Eikhof, G., Na currents and action potentials in rat myeli-nated nerve fibres at 20 and 37°C, Pflügers Arch., 409, 569–577, 1987.

30. Sweeney, J.D., Mortimer, J.T., and Durand, D., Modeling of mammalian my-elinated nerve for functional neuromuscular electrostimulation, in IEEE 9thAnnu. Conf. Eng. Med. Biol. Soc., Boston, MA, 1987, pp. 1577–1578.

31. Rattay, F., Simulation of artificial neural reactions produced with electricfields, Simulation Practice Theory, 1, 137–152, 1993.

32. Rattay, F. and Aberham, M., Modeling axon membranes for functional elec-trical stimulation, IEEE Trans. Biomed. Eng., BME 40, 1201–1209, 1993.

33. Schwarz, J.R., Reid, G., and Bostock, H., Action potentials and membranecurrents in the human node of Ranvier, Eur. J. Physiol., 430, 283–292, 1995.

34. Wesselink, W.A., Holsheimer, J., and Boom, H.B., A model of the electricalbehaviour of myelinated sensory nerve fibres based on human data, Med.Biol. Eng. Comput., 37, 228–235, 1999.

35. Rattay, F., Lutter, P., and Felix, H., A model of the electrically excited humancochlear neuron. I. Contribution of neural substructures to the generation andpropagation of spikes, Hear. Res., 153, 43–63, 2001.

36. Burke, D., Kiernan, M.C., and Bostock, H., Excitability of human axons, Clin.Neurophysiol., 112, 1575–1585, 2001.

37. Baker, M.D., Axonal flip-flops and oscillators, Trends Neurosci., 23, 514–519,2000.

38. Magee, J.C., Voltage-gated ion channels in dendrites, in Dendrites, Stuart, G.,Spruston, N., and Häusser, M., Eds., Oxford University Press, London, 1999,pp. 139–160.

39. Belluzzi, O. and Sacchi, O., A five conductance model of the action potentialin the rat sympathetic neurone, Progr. Biophys. Molec. Biol., 55, 1–30, 1991.

40. Winslow, R.L. and Knapp, A.G., Dynamic models of the retinal horizontalcell network, Prog. Biophys. Mol. Biol., 56, 107–133, 1991.

41. McCormick, D.A. and Huguenard, J.R., A model of the electrophysiologicalproperties of thalamocortical relay neurons, J. Neurophysiol., 68, 1384–1400,1992.



42. Traub, R.D., Jefferys, J.G., Miles, R., Whittington, M.A., and Toth, K., A branch-ing dendritic model of a rodent CA3 pyramidal neurone, J. Physiol., 481, 79–95,1994.

43. DeSchutter, E. and Smolen, P., Calcium dynamics in large neuronal models,in Methods in Neuronal Modeling: From Ions to Networks, 2nd ed., Koch, C. andSegev, I., Eds., MIT Press, Cambridge, MA, 1999, pp. 211–250.

44. Fohlmeister, J.F. and Miller, R.F., Impulse encoding mechanisms of ganglioncells in the tiger salamander retina, J. Neurophysiol., 78, 1935–1947, 1997.

45. Fohlmeister, J.F. and Miller, R.F., Mechanisms by which cell geometry controlsrepetitive impulse firing in retinal ganglion cells, J. Neurophysiol., 78,1948–1964, 1997.

46. Hartmann, R., Topp, G., and Klinke, R., Discharge patterns of cat primaryauditory fibers with electrical stimulation of the cochlea, Hearing Res., 13,47–62, 1984.

47. Motz, H. and Rattay, F., A study of the application of the Hodgkin–Huxleyand the Frankenhaeuser–Huxley model for electrostimulation of the acousticnerve, Neuroscience, 18, 699–712, 1986.

48. Hodgkin, A.L. and Katz, B., The effect of temperature on the electrical activityof the giant axon of the squid, J. Physiol., 109, 240–249, 1949.

49. Huxley, A.F., Ion movements during nerve activity, Ann. N.Y. Acad. Sci., 81, 221–246, 1959.

50. Rattay, F., Propagation and distribution of neural signals: a modeling studyof axonal transport, Physics Alive, 3, 60–66, 1995.

51. Frankenhaeuser, B. and Huxley, A.L., The action potential in the myelinatednerve fibre of Xenopus laevis as computed on the basis of voltage clamp data,J. Physiol., 171, 302–315, 1964.

52. Frankenhaeuser, B. and Moore, L.E., The effect of temperature on the sodiumand potassium permeability changes in myelinated nerve fibers of Xenopuslaevis, J. Physiol., 169, 431–437, 1963.

53. Halter, J.A. and Clark, J.W., A distributed-parameter model of the myelinatednerve fiber, J. Theor. Biol., 148, 345–382, 1991.

54. Ritchie, J.M., Physiology of axons, in The Axon: Structure, Function and Patho-pysiology, Waxman, S.G., Kocsis, J.D., and Stys, P.K., Eds., Oxford UniversityPress, Oxford, 1995, pp. 68–96.

55. Ranck, J.B., Jr., Which elements are excited in electrical stimulation of mam-malian central nervous system: a review, Brain Res., 98, 417–440, 1975.

56. Warman, E.N., Grill, W.M., and Durand, D., Modeling the effects of electricfields on nerve fibers: determination of excitation thresholds, IEEE Trans.Biomed. Eng., 39, 1244–1254, 1992.

57. Plonsey, R. and Barr, R.C., Electric field stimulation of excitable tissue, IEEETrans. Biomed. Eng., 42, 329–336, 1995.

58. Zierhofer, C.M., Analysis of a linear model for electrical stimulation of axons:critical remarks on the “activating function concept,” IEEE Trans. Biomed. Eng.,48, 173–184, 2001.

59. Rattay, F., Minassian, K., and Dimitrijevic, M. R., Epidural electrical stimula-tion of posterior structures of the human lumbosacral cord: 2. Quantitativeanalysis by computer modeling, Spinal Cord, 38, 473–489, 2000.

60. Henriquez, C.S., Simulating the electrical behavior of cardiac tissue using thebidomain model, Crit. Rev. Biomed. Eng., 21, 1–77, 1993.



61. Knisley, S.B., Trayanova, N., and Aguel, F., Roles of electric field and fiberstructure in cardiac electric stimulation, Biophys. J., 77, 1404–1417, 1999.

62. Van den Honert, C. and Stypulkowski, P.H., Temporal response patterns ofsingle auditory nerve fibers elicited by periodic electrical stimuli, Hear. Res.,29, 207–222, 1987.

63. Javel, E. and Shepherd, R.K., Electrical stimulation of the auditory nerve: II.Effect of stimulus waveshape on single fibre response properties, Hear. Res.,130, 171–188, 1999.

64. Horikawa, Y., Simulation study on effects of channel noise on differentialconduction at an axon branch, Biophys. J., 65, 680–686, 1993.

65. Holden, A.V., Models of the Stochastic Activity of Neurons, Springer-Verlag,Berlin, 1976.

66. Sigworth, F.J., The variance of sodium current fluctuations at the node ofRanvier, J. Physiol., 307, 97–129, 1980.

67. Verveen, A.A. and Derksen, H.E., Fluctuation phenomena in nerve mem-brane, Proc. IEEE, 56, 906–916, 1968.

68. DeFelice, L.J., Introduction to Membrane Noise, Plenum Press, New York, 1981.69. Rubinstein, J.T., Threshold fluctuations in an N sodium channel model of the

node of Ranvier, Biophys. J., 68, 779–785, 1995.70. Rattay, F., Basics of hearing theory and noise in cochlear implants, Chaos,

Solitons Fractals, 11, 1875–1884, 2000.71. Greenberg, R.J., Velte, T.J., Humayun, M.S., Scarlatis, G.N., de Juan, E., Jr., A

computational model of electrical stimulation of the retinal ganglion cell, IEEETrans. Biomed. Eng., 46, 505–514, 1999.

72. Toris, C.B., Eiesland, J.L., and Miller, R.F., Morphology of ganglion cells inthe neotenous tiger salamander retina, J. Comp. Neurol., 352, 535–559, 1995.

73. Velte, T.J. and Miller, R.F., Dendritic integration in ganglion cells of the mud-puppy retina, Visual. Neurosci., 12, 165–175, 1995.

74. Coleman, P.A. and Miller, R.F., Measurement of passive membrane parame-ters with whole-cell recording from neurons in the intact amphibian retina,J. Neurophysiol., 61, 218–230, 1989.

75. Fohlmeister, J.F., Coleman, P.A., and Miller, R.F., Modeling the repetitive firingof retinal ganglion cells, Brain Res., 510, 343–345, 1990.

76. Geddes, L.A. and Baker, L.E., The specific resistance of biological material-Acompendium of data for the biomedical engineer and physiologist, Med. Biol.Eng., 5, 271–293, 1967.

77. Lukasiewicz, P. and Werblin, F., A slowly inactivating potassium currenttrunkates spike activity in ganglion cells of the tiger salamander retina, J.Neurosci., 8, 4470–4481, 1988.

78. Lipton, S.A. and Tauck, D.L., Voltage-dependent conductances of solitaryganglion cells dissociated from rat retina, J. Physiol., 385, 361–391, 1987.

79. Hines, M.L. and Carnevale, N.T., NEURON: a tool for neuroscientists, Neu-roscientist, 7, 123–135, 2001.

80. Wiley, J.D. and Webster, J.G., Analysis and control of the current distributionunder circular dispersive electrodes, IEEE Trans. Biomed. Eng., BME-29,381–385, 1982.

81. Humayun, M.S., de Juan, E., Dagnelie, G., Greenberg, R.J., Propst, R., andPhillips, D.H., Visual perception elicited by electrical stimulation of retina inblind humans, Arch. Ophthalmol., 114, 40–46, 1996.



82. DeSchutter, E. and Bower, J.M., An active membrane model of the cerebellarPurkinje cell: I. Simulation of current clamps in slice, J. Neurophysiol., 71,375–400, 1994.

83. Rijkhoff, N.J., Holsheimer, J., Koldewijn, E.L., Struijk, J.J., Van Kerrebroeck,P.E., Debruyne, F.M., and Wijkstra, H., Selective stimulation of sacral nerveroots for bladder control: a study by computer modeling, IEEE Trans. Biomed.Eng., 41, 413–424, 1994.

84. Bugbee, M., Donaldson, N.N., Lickel, A., Rijkhoff, N.J., and Taylor, J., Animplant for chronic selective stimulation of nerves, Med. Eng. Phys., 23, 29–36,2001.

85. Rattay, F., Modeling the excitation of fibers under surface electrodes, IEEE-Trans. Biomed. Eng., BME-35, 199–202, 1988.

86. Coburn, B. and Sin, W.K., A theoretical study of epidural electrical stimulationof the spinal cord. Part I: Finite element analysis of stimulus fields, IEEE Trans.Biomed. Eng., 32, 971–977, 1985.

87. Rattay, F., Leao, R.N., and Felix, H., A model of the electrically excited humancochlear neuron. II. Influence of the three-dimensional cochlear structure onneural excitability, Hear. Res., 153, 64–79, 2001.

88. Frijns, J.H., de Snoo, S.L., and Schoonhoven, R., Improving the accuracy ofthe boundary element method by the use of second-order interpolation func-tions, IEEE Trans. Biomed. Eng., 47, 1336–1346, 2000.

89. Bower, J. and Beeman, D., The Book of GENESIS, 2nd ed., TELOS, New York,1997.

90. Mascagni, M.V. and Sherman, A., Numerical methods for neural modeling,in Methods in Neuronal Modeling: From Ions to Networks, 2nd ed., Koch, C. andSegev, I., Eds., MIT Press, Cambridge, MA, 1999, pp. 569–606.




section three

Stimulating and recording of nerves and neurons


chapter four

Stimulating neural activity

James D. Weiland, Mark S. Humayun, Wentai Liu, and Robert J. Greenberg

Contents

4.1. Introduction4.2 Basic theory and methods

4.2.1 Membrane biophysics of neurons4.2.2 Electrical activation of voltage-gated ion channels

4.3 Achieving desired response with stimulus waveform manipulation4.3.1 Electrode shape and position dependence

4.4 Stimulus pulse dependence4.4.1 Charge balancing and cell damage4.4.2 Selective stimulation of cells

4.5 Practical examples and applicationsReferences

4.1. IntroductionThe study of interactions between electricity and biological tissue has been anactive research area for over 100 years. The use of electrical stimulation as aneffective therapy to treat human diseases has only recently been realized,thanks in equal parts to advances in electronics, neuroscience, and medicine.In the 18th century, Allesandro Volta and Luigi Galvani both investigatedsensory and muscular responses to electrical stimulation. Galvani stimulatedthe muscles of frogs, while the intrepid Volta was his own test subject, placingelectrodes in either ear and reporting a “crackling sound” when power wasapplied. Further study of nerve and muscle identified these tissues as excitable(using electrical signals to communicate) and resulted in more careful inves-tigations of electrical stimulation as a potential therapy.



Among the pioneers in the field of neural prostheses were Djourno andEyries,1 who reported the first cochlear implants, and Brindley,2 whoimplanted an electrical stimulation device on the cortex of a blind test subject.When Djourno and Eyres implanted four patients with a simple device, asingle wire coupled to a sound transducer, they were able to distinguishseveral distinct sounds but could not understand speech. House advancedthe cochlear implant by implanting stimulation devices in the cochlea andcochlear nucleus.3 Brindley implanted a blind test subject with a devicecapable of stimulating the brain at 80 locations, and Dobelle has implanteda similar device in humans. One of Dobelle’s patients has been implantedfor over 20 years and has demonstrated the ability to identify 6-inch lettersat a distance of 5 feet, corresponding to 20/120 vision.4,5 These early resultsin humans demonstrated the potential of electrical stimulation as a meansto replace lost sensory function and motivated a growing area of researchinto stimulation devices and physiology of electrically stimulated muscleand nerve.

This chapter describes methods for measuring the neural response toelectrical stimulation. A brief review of membrane biophysics is presentedto establish the mechanism by which electric current can mimic the naturalneural response. Several analysis methods are described for biopotentialrecording, and alternatives to biopotential recording are discussed. Alsopresented is a review of experiments that demonstrate manipulation of thestimulus site by control of the waveform. Limits of stimulus waveforms arepresented, as well. The chapter concludes with examples of implementationof the principles in commercial neural prostheses.

4.2 Basic theory and methodsAlthough sometimes considered an elemental building block, the neuronitself is a sophisticated biological system with a variety of subcellular struc-tures designed to maintain a chemical and electrical equilibrium across thecell membrane; respond to chemical, electric, mechanical, or light stimuli;and signal messages to other neurons. The excitable membrane of the neuronhas molecular sensors that respond to a variety of stimuli. If the responsechanges the electrical characteristics of the cell membrane, then the electricalpotential across the cell (the membrane potential) will change. Signals aretransmitted through nerve cells by temporary, controlled fluctuations in themembrane potential.

4.2.1 Membrane biophysics of neurons

Membrane biophyiscs is a broad area of research to which entire textbookshave been devoted. The discussion below is derived mainly from twosources: Bioelectricity: A Quantitative Approach, by Plonsey and Barr,6 andPrinciples of Neural Science, 3rd ed., by Kandel et al.7 Nerve cells have ameasurable electrical potential that exists across the cell membrane. Typical



resting potential is –60 mV measured inside with reference to outside. Theelectrical potential exists due to active and passive cell processes that main-tain a difference in the chemical concentration of various ions that exist inthe intra- and extracellular fluids.

The ions most responsible for maintenance of membrane potential areNa+, Cl–, and K+, although in some cells (cardiac and photoreceptor) Ca2+

plays a major role in signaling. The concentrations of these ions are differentinside and outside the cell. Potassium (K+) has a higher concentration insidethe cell, and sodium (Na+) and chloride (Cl–) have higher extracellular con-centrations. Table 4.1 shows the concentrations in the squid axon.6

In human nerve cells, the absolute numbers are two to three times lower,but the ratio of intracellular to extracellular is similar. The cell membrane atrest is not equally permeable to all ions; that is, some ions diffuse morereadily than others. The diffusion force will drive the ions toward equalconcentration inside and outside the cell. However, taking an extreme exam-ple, if only positively charged ions could cross a membrane and negativelycharged ions were completely blocked from diffusing, the diffusion of onlypositive charge would create an electrical imbalance that would slow andthen halt the diffusion. The potential at which the diffusion force is equaland opposite to the electrical force is called the Nernst potential and can becalculated from Eq. (4.1),

(4.1)

where Vmp is the membrane voltage for the pth ion; Zp is the charge of thepth ion, [Cp]e is the extracellular concentration of the pth ion, and [Cp]i is theintracellular concentration of the pth ion. The Nernst potential is for a singleion. The various ions in the intra- and extracellular fluid each have Nernstpotentials. The resting permeabilities and Nernst potentials for each ion arelisted in Table 4.1. By weighting the Nernst potential of each ion with themembrane permeability for that ion, it is possible to calculate the membranepotential at steady state using the Goldman equation:

(4.2)

Table 4.1 Concentration, Permeability, and Nernst Potential for Various Ions in Squid Axon

Ion

Concentration (mM/l) Relative PermeabilityNernst

PotentialIntracellular Extracellular At RestFor Active Membrane

K+ 280 10 1.0 1.0 –83.9Na+ 61 485 0.04 20.0 52.2Cl– 51 485 0.45 0.45 –56.7

V Z C C mVmp p p e p i= 58 10/[ log ([ ] /[ ] )]

V RT F P K P Na

P Cl P K P Na P Cl

m K e Na e

Cl i K i Na i Cl e

= +

+ + +

+ +

− + + −

( / ) ln(( [ ] [ ]

[ ] )/( [ ] [ ] [ ] ))



where Px is the permeability of the membrane to ion x. Though this equationis only valid if the membrane is at rest (net current is zero), qualitatively itis clear that a change in permeability for one or more ions will alter themembrane potential.

Electrical charge moves across the cell membrane through ion channelsand ion pumps. The rest of the cell membrane is impermeable to ion flow.The number and nature of ion channels in the membrane determine thepermeability of the membrane. Ion channels can always be open or can beopened and closed in response to a stimulus, but once they are open theyact as pores that allow ions to flow to equilibrium. Ion channels can bespecific to particular ions but are otherwise passive; that is, the chemical orelectrical driving forces will determine the direction of the net ion flowthrough the channel. In contrast, ion pumps move ions against the drivingforce gradient, consuming energy in the process.

Signals are initiated by changes in the membrane permeability, whichchange the membrane potential (see Eq. (4.2)). Membrane permeability canbe changed in a variety of ways depending on the type of cell. If the cell isa sensory transducer, such as a photoreceptor in the retina, then the presenceof the excitation signal results in a change of membrane potential. Thephotoreceptors are unusual in that they are depolarized (excited) in theabsence of light. Other sensory receptors, such as the hair cells of the cochleaand olfactory cells of the nose, are depolarized in response to their excitation,either sound or odor. The excitation signal might trigger a chemical cascadethat opens ion channels or the ion channels may be mechanically opened.The second way that membrane potential can be modified is by synaptictransmission, that is, through a signal transmitted from a connecting neuron.Chemical neurotransmitters released by a nearby neuron bind to a specificprotein on the membrane of the target neuron, leading to the opening ofchemically gated ion channels.

The stimulus that creates a change in the membrane permeability affectsthe equilibrium equation and thus alters the cell membrane potential, whichtriggers the opening of a special type of ion channel that is sensitive totransmembrane potential. These ion channels are called voltage-gated ionchannels, and depolarization changes membrane permeability by openingvoltage-gated channels. Initially, the stimulus will result in a proportionalresponse from the cell. In this case, the cell membrane will respond withwhat is called a graded potential. This means that there is a proportionalrelationship between the amount of depolarization and the strength of stim-ulus. Graded potential cells, such as the bipolar cells in the retina, transmitinformation over short distances; the space constant for signal propogationis small (the signal will decay over distance). Some cells have the ability to“spike,” to produce an action potential, a self-propogating signal that trans-mits over great lengths. The anatomical structure that serves as the trans-mission line for the action potential is the cell axon, a relatively long projec-tion from the cell body (soma). The action potential is generated when thecell depolarization exceeds a threshold. Once a cell is depolarized beyond



that point, a process begins that will generate an action potential, even if thestimulus that produces the initial depolarization is removed.

Hodgkin and Katz determined the permeability of the squid axon mem-brane to sodium, potassium, and chloride ions, both at rest and at the peakof the action potential. They obtained the data by altering the extracellularion concentrations and monitoring the current. The results suggested thatthe action potential generation is due to a dramatic change in the perme-ability of the membrane to sodium. The permeabilities from this work areshown in Table 4.1.

This sequence of events that occurs during an action potential wasdescribed in detail by Hodgkin and Huxley,1 who won the Nobel Prize fortheir seminal work in membrane biophysics. The initial depolarization(from synaptic transmission or sensory stimuli) will cause voltage-gatedion channels to activate (open). Voltage-gated ion channels exist for bothsodium and potassium, but the sodium channels have a much shorter timeconstant. The quick response from the sodium channels results in anincrease in the outward current, which depolarizes the membrane evenfurther. The ion channels cannot open or close instantaneously, so, onceopened, they will stay open for a finite period of time. This allows thedepolarization to continue even after the stimulus is removed. An actionpotential is generated when the depolarization exceeds a threshold, atwhich point the outward current exceeds the inward current, creating apositive feedback situation (i.e., more outward current, more depolariza-tion, more open sodium channels). The depolarization is quickly (within1 to 2 msec) reversed by two mechanisms: inactivation (closing) of voltage-gated sodium channels and activation of voltage-gated potassium chan-nels. Sodium activation and sodium inactivation are accomplished by twodistinct, voltage-sensitive mechanisms. The sodium inactivation mecha-nism and the potassium activation mechanism have longer time constantsthan sodium activation, but they work together to restore membrane poten-tial to resting potential.

After the action potential, it takes several milliseconds for the voltage-gated ion channels to return to their resting state. The voltage-gated sodiumchannels will change from inactive (cannot be opened) to resting (closed,but will open in response to voltage change). The potassium channels willchange from active (open, thereby making the membrane highly permeableto inward potassium current) to resting (closed, but will open in responseto depolarization). This transitional period is called the refractory period andhas two stages. The absolute refractory period is the time during which theincreased potassium permeability prevents an action potential generationregardless of the stimulus. The relative refractory period follows, duringwhich time an increased stimulus is needed to generate an action potential.Hodgkin and Huxley quantified the behavior of sodium and potassiumcurrents with mathematical equations. This work, and that of Frankenhauserand Huxley, remains the basis for much of the neuron modeling that isperformed today.



The action potential has been described as an all-or-nothing phenome-non, implying that information on stimulus strength cannot be communi-cated by these cells. However, the firing rate of the neuron, the number ofspikes per second, can be used to code information about stimulus strength.In addition, it has been demonstrated in the retina that cells may work inconcert to transmit more information than if they worked independently

In summary, the basic steps of transient ion channel response to depo-larization are: (1) initial depolarization; (2) voltage-gated sodium channelactivation, leading to more depolarization; (3) more voltage-gated sodiumchannels being open due to continued depolarization and net outward cur-rent; (4) inactivation of sodium channels and activation of potassium chan-nels; (5) repolarization of cell membrane; and (6) ion channels return to theresting state. Ion channels are sophisticated structures built from transmem-brane proteins. The study of ion channel membrane biophysics is a fieldunto itself and a detailed discussion is beyond the scope of this chapter.While the description above is a generalization of membrane behavior todemonstrate how the neuron is capable of signaling by transiently alteringits membrane properties, the basic mechanism of action potential generationis shared by most neurons.

4.2.2 Electrical activation of voltage-gated ion channels

Electrical energy applied to the neuron has the ability to elicit a gradedpotential and action potential from the neuron, thereby allowing electricalstimulation to replace lost function in the sensory or motor systems. Apply-ing a negative current via an extracellular stimulating electrode has the effectof depolarizing the membrane.9 The membrane depolarization from theexternal, artificial stimulus initiates the natural processes of transient mem-brane depolarization. A sufficient electrical stimulus can lead to action poten-tial generation.

The stimulating electrode typically serves as the cathode because nega-tive current leads to membrane depolarization. However, because a currentgenerator has a source and a sink, a return electrode is required to completethe current loop through the tissue. Monopolar stimulation occurs when thereturn electrode is large and relatively remote from the stimulating electrode.Because the return electrode is large, the density of the current at this site issmall so neuron activation only occurs at a single site, the monopole. If thereturn electrode is similar in size to the stimulating electrode, then the currentdensity at the return is similar to the stimulation, allowing for neuron acti-vation to occur at both electrodes in the pair, although such activation is lesslikely at the anode as anodic current is less effective at producing a response.This configuration is referred to as a dipole and has the advantage of reducingthe spread of electrical current.

Mathematical models of electrical stimulation suggest that current den-sity, not current, is the important parameter for calculating membranepotential, but experimentally this principle does not always hold. Current



density is directly proportional to the electric field created by a stimulatingelectrode, and the electric field is frequently used as the input to modelsof stimulation. It follows that a reduction in the electrode size should beaccompanied by a proportional reduction in stimulus current, but experi-mentally this is not true over all ranges of electrode sizes. A study usingskin electrodes of varying diameter over several orders of magnitude foundthat response threshold was independent of diameter below a certain diam-eter but does become dependent on diameter as diameter increases.10 Theskin stimulation study hypothesized that current was conducted througha number of discrete channels through the skin, and the effective currentdensity was dependent on the number of channels activated. Hence, theeffective current density could be different from the calculated currentdensity, the latter assuming uniform conduction through an isotropicmedia. Mathematical models of neural activation with extracellular elec-trodes use the stimulating current or current density as input, but thesemodels typically assume an ideal electrode geometry (point source orsphere) in an isotropic conducting media.11,12 The more complex situationof current flow between two electrodes in an anisotropic media will requirenumerical techniques to corroborate experimental data.

Electrical activation of neurons can also be achieved through intracellu-lar stimulation. The technique used to inject current inside the cell is thepatch clamp.13,14 Glass pipette electrodes are used to penetrate the cell mem-brane. A high-impedance seal that forms between the membrane and pipetteprevents stimulus current from leaking out the hole formed by the pipette.Small currents (nA) significantly affect membrane potential. Although thisis an extremely efficient means of stimulating the cell in terms of how muchenergy is needed to produce an action potential, it would be difficult torealize an implantable device with intracellular electrodes, so this approachwill not be considered further.

Electrical stimulus thresholds (i.e., the amount of electrical stimulusrequired to produce an action potential) are sometimes defined in coulombs(C), a basic unit of electrical charge (1 C equals the charge of 6

× 1018

electrons).15 The more familiar electrical current unit, amperes (A), is chargeper unit time, or C/sec. The amount of charge in a rectangular stimuluspulse is simply the product of the stimulus strength in amperes and thestimulus duration in seconds. For more complex pulse shapes, the stimuluscharge can be determined from the area under the pulse. Because a stimuluspulse is typically less than 1 msec and stimulus current less than 1 mA,microcoulombs (

µC) or nanocoulombs (nC) are used to quantify the stimuluscharge threshold. One reason for using stimulus charge to define threshold,as opposed to stimulus current, is that current will change dramatically withthe duration of the pulse, but stimulus charge, though not constant, willchange less. Therefore, using charge allows the stimulus threshold to bestated as a single parameter. If current is used to define the threshold, thepulse duration must also be included. The best practice may be to presentresults with all of the pertinent parameters.



The relationship between stimulus current threshold and stimulusduration is often expressed graphically with a strength–duration curve.6Stimulus threshold current is plotted vs. stimulus pulse duration over atleast one order of magnitude. Parametric models can be created from theempirically determined strength–duration curves to characterize theresponse of a cell to electrical stimulus over a range of pulse duration.These curves can also be used to quantify the rheobase current and thechronaxie of the cell. Rheobase is the minimum current required to stim-ulate a cell, regardless of the duration of the stimulus pulse. Chronaxie isthe duration of the stimulus pulse required when twice the rheobase cur-rent is used. A shorter chronaxie implies that the cell will respond withless stimulus charge at that current level; that is, the cell is more sensitive.Figure 4.1 shows a simulated strength–duration curve. As stimulus pulseduration increases, threshold current decreases but threshold chargeincreases. Also, note the region of near-linear stimulus charge with shorterpulses. This demonstrates the value of threshold charge as a parameter todefine the properties of the cell. For stimulus pulse durations from 0.05 to0.5 msec, the threshold charge changes only 10% even though the pulseduration changes one order of magnitude. Shorter pulses are preferred, asthey require less total energy.

A second method of investigating electrical response is the ampli-tude–intensity function. This plots the stimulus strength at a fixed-pulseduration vs. the amplitude of the electrically elicited response.16 The responsein this case is an evoked potential or summed response of many neurons.This typically yields an increasing line that will saturate at some stimuluslevel (Figure 4.2). One advantage of the input–output function is that itallows a predication of true threshold by extrapolation of the input–outputfunction to the intersection of the x-axis. The strength–duration curve thresh-old is affected by noise and sometimes determined subjectively.

Figure 4.1 Simulated strength–duration curve. The rheobase current is the minimumcurrent required to stimulate the neuron regardless of pulse duration. Chronaxie isthe pulse duration corresponding to twice the rheobase current. At shorter pulseduration, the curve is nearly linear.



Evoked potentials are produced by averaging many responses to thesame stimulus. Averaging N trials will improve the signal-to-noise ratioby

(100 averages will increase the signal-to-noise ratio by 10).17 Thisnoise reduction strategy cannot be applied to studies of single units (actionpotentials from a single cell) because the action potential is so short (<1msec) and, due to the stochastic nature of action potential firing, can beoffset by 1 msec between trials. An offset of 1 msec will result in dimin-ishing by half the signal if two trials are averaged. To overcome this, singleunits are treated as discrete events occurring at a specific time. Over manytrials, it is possible to accumulate spike event times after an electricalstimulus (or any stimulus). These spike times are then grouped into timewindows called bins; for example, a 1-sec recording can be divided into100 10-msec bins. The results of this sorting are displayed in a post-stimulus time histogram (PSTH). This is a straightforward procedure ifonly one cell is responding with the same action potential each time andthe action potential signal is well above the noise, allowing a simplethreshold rule for event detection. The PSTH allows us to demonstratethat the single unit firing is correlated to the stimulus if a particular binrises above the spontaneous firing rate. However, in most instances theanalysis becomes more difficult. The action potential may not be signifi-cantly above the noise, or multiple action potentials from multiple cellsmay be recorded simultaneously. Signal processing strategies have beendeveloped to attack these problems, but a thorough review is beyond thescope of this chapter.

Strength–duration curves, amplitude–intensity functions, and PSTH aretime-honored methods of biopotential analysis. These have been and willcontinue to be the first and best measures of the response of neural tissuesto electrical stimulation. They are easily interpreted and can be acquiredwith relatively simple, inexpensive equipment; however, biopotential anal-ysis is not without its limitations, particularly when used to assess electricallyelicited neural response. First and foremost is the problem of the electricalstimulus artifact.

Figure 4.2 Simulated input–output function. The response amplitude saturates as thestimulus increases. It is possible to estimate the true threshold by extrapolation ofthe function to the x-axis intercept.

N



The electric field created in the tissue by the electrical stimulus cansaturate amplifiers and mask an electrical response. This is especially prob-lematic when the recording electrodes are positioned close to the stimuluselectrodes. Potential solutions for artifact reduction have been discussed byParsa et al.18 Simple methods to reduce the artifact include changing thebandpass of the amplifier to allow for a quicker recovery, using a sampleand hold amplifier to prevent saturation, and using differential recordingwith the reference electrode on an equipotential line with the recordingelectrode. Methods for avoiding artifact include recording “downstream” orat the proximal neural areas. An example would be recording in the inferiorcolliculus and stimulating in the auditory nerve. A delay of several millisec-onds between stimulus and response would allow the artifact to subsidebefore the response is present.

A second shortcoming of biopotential recording is the limited amountof tissue that is observed. The response is only recorded at a single point,or multiple points if several electrodes are used. This requires that the record-ing electrode be positioned such that it will record the response from thecell being activated by the stimulating electrode. The presence of interneu-rons can confound results.

To augment data from biopotential recording, neural responses can berecorded with a variety of optical and imaging methods. Voltage-sensitivefluorescent dyes reflect the actual membrane potential, while optical intrin-sic signaling relies on changes in optical properties of tissue related toincreased metabolic activity. Voltage-sensitive dyes (VSDs) are frequentlyused in brain slice recording.19 This preparation lends itself to voltage-sensitive dyes, as the dissected tissue can easily be processed to load thedye in the cells. Further, the thickness of the tissue, which will affect thedye penetration, is controllable. It is more difficult to use dyes to penetrateintact tissue, such as an isolated retina, which may have some barriers todiffusion (note that a retinal slice is different from an isolated retina). Ithas been demonstrated that VSDs reflect the activity in the dendriticpostsynatpic area of the neuron; that is, VSDs are markers of subthresholdactivity. It is possible to combine VSD imaging with electrical recording tomap both subthreshold and action potential activity.20

Optical intrinsic signaling has been used to study tissue activity for sometime. In general, optical imaging does not require the use of fluorescent dyesor other extrinsic markers. The basic premise is that neuronal activity causesa transient increase in cell volume or blood flow that results in a change inlight scatter or absorption. Optical imaging can be used to study activity onthe surface of the brain or in a dissected section of brain tissue, such as theretina. Optical imaging has also been used to study electrically evokedchanges in activity in isolated brain tissue slices. Kohn et al.21 used threetrains of electrical pulses of frequencies 10, 2, and 1 Hz. They found that itwas possible to observe an increase in the transmitted light resulting fromstimulation. The increase in transmitted light is most likely due to transientcellular swelling. The electrical activity of the stimulated neurons causes Na+



and Cl– to be drawn into the cell. The increase in cellular volume is due towater drawn into the cell to equilibrate the osmotic pressure across the cellmembrane. As the cellular volume increases, the concentration of scatteringparticles within the cell decreases, causing the light scatter of the cell todecrease. The reduced light scatter allows more light to be transmitted bythe brain tissue slice. If the imaging system was measuring reflected lightinstead of transmitted light, the image intensity would likely decrease, asthe reflected light is directly related to the light scatter.

Optical imaging of tissue requires the use of sensitive equipment toobtain the image. The changes in optical properties are small and requireexpensive imaging chips to detect these changes above the noise of thecamera. It must be possible to visualize the tissue that is being analyzed. Ifcortical tissue is being analyzed, then the skull must be removed over thearea of interest.

Functional magnetic resonance imaging (fMRI) is used to detectincreased brain activity in response to specific stimuli. It has the advantageof being truly noninvasive. No direct electrical interface is required nor isdirect visualization of the neural tissue necessary. Preliminary experimentsindicate that the electrical stimulus will result in a stimulus artifact thatobscures any measurable response. To counter this, it is possible to recordthe scan after the electrical stimulus has ceased. Even without this limitation,fMRI would be limited to spatial information. Also, the spatial resolution islow, typically 1 mm or greater. Finally, fMRI is difficult to perform, requiringexpensive equipment. This type of data may supplement electrical responses,but probably would be of limited value alone.

The biochemistry of the neuron is altered by continuous increased activity,and these changes can be detected with histological markers. These methodsrequire that the experimental animal be sacrificed a short time after the stim-ulus and the tissue processed appropriately. One such marker is the proteinCfos, which is transiently expressed 1 to 4 hours after neural excitation.22 Thisprotein can be marked with antibodies and has been used to investigateelectrical stimulation of the cochlea and light responses in the retina.23–25 Ingeneral, it can be used to indicate the extent of excitation for a given stimulus.For example, in the Saito study,24 the cochlea was electrically activated in anarea known to respond to sound waves in a particular frequency range. His-tological evaluation of the dorsal cochlear nucleus and inferior colliculusshowed increased Cfos expresssion in a narrow band, indicating that only thatfrequency band of the auditory system was activated. However, Cfos cannotdistinguish between primary and secondary excitation, so unless the neuralstructure is anatomically well defined, such as the tonotopic lamina of theinferior colliculus or the ocular dominance columns of the striate cortex, thenCfos marking will be diffuse. Also, only one experimental condition can betested because the animal must be sacrificed shortly after testing to processthe tissue. Finally, the antibody is species dependent, so while the methodmay work well in one animal, it may not work in another if the Cfos antibodyin the second animal is not well characterized.



4.3 Achieving desired response with stimulus waveform manipulation

4.3.1 Electrode shape and position dependence

The shape and position of the stimulating electrode affect the electricalresponse of neural tissue because the electric field created by the stimulatingcurrent is affected by these same electrode parameters. The effects of shapeare most pronounced when the electrode is in close proximity to the neurons.Planar electrodes have a nonuniform current density that is significantlyhigher at the edges than in the center of the electrode.26 The current densityat a circular electrode within a nonconducting surface has an analyticalsolution that predicts that the current density at the edge of the electrodewill be infinite. In practice, such effects are seen as skin burns at the perimeterof circular defribrillators. This may affect the response by selectively stimu-lating neurons in these areas of higher current density where the desiredoutcome would be to equally stimulate all neurons near the electrode. Thenonuniform electric field of the planar electrode can be moderated by recess-ing the electrode surface in the nonconducting substrate or by changing theshape of the electrode itself to a half sphere.27

The position of the electrode will affect how well a response can betargeted to a small area of tissue. Due to current spread, an electrode farfrom the desired area will stimulate a larger area then a closer electrode.However, the current density is decreased with current spread, so morecurrent is required to stimulate when the electrode is positioned farther awayfrom the target neurons. In spite of these facts, optimal electrode positionmust balance anatomical considerations against proximity to neurons toarrive at the best solution. In the case of positioning a stimulating electrodeon the retina, an electrode on the surface of the retina is less disruptive tothe anatomy than an electrode that would penetrate into the retina. Theelectrical properties of electrodes implanted in the cortex are known to besignificantly affected by the foreign-body reaction that surrounds the elec-trode. The reactive tissue has an insulating effect that degrades the electricalcontact between the tissue and the electrode.

4.4 Stimulus pulse dependence4.4.1 Charge balancing and cell damage

The electrode position and shape are analogous to computer hardware;that is, they are physically defined parameters that are impossible tochange once set in place. What can be changed is the software, or, in thecase of an electrical stimulating device, the stimulus pulse. It is possibleto target specific neural structures by varying the stimulus pulse param-eters. However, some general rules must be followed when electricallystimulating neural tissue with metal electrodes. First is the concept of



charge balanced stimulation, first reported by Lilly in 1961.28 For a singlestimulation waveform, the total net charge must be zero. This can beaccomplished either by supplying equal cathodic and anodic current fromthe stimulator or by a blocking capacitor that will slowly discharge aftera monopolar stimulus pulse. If charge balance is not obtained, then, overtime, the net charge accumulation on the electrode may increase the elec-trode potential to the point where harmful quantities of gaseous oxygenor hydrogen are produced (bubbling). In practice, stimulators are designedto have blocking capacitors. The size of the capacitor must be chosencarefully, as a capacitor that is too small will severely limit the durationof the pulse that can be used.

A charge-balanced waveform is not necessarily a safe waveform. Withinthe course of a stimulus pulse that is overall charge balanced, it is possibleto momentarily exceed empirically established safety limits for electrodepotential, charge density, or total charge. Two safety limits should be con-sidered: neural damage limits and electrochemical limits. Neural damagelimits are dependent on the ability of biological tissue to withstand theelectric current without degrading. Electrochemical limits are based on theability of the electrode to store or dissipate electrical charge without exceed-ing the water window, which is the potential window outside of which sig-nificant bubble formation is evident at the interface.

Neural damage limits have been studied extensively by Pudenz, Agnew,and McCreery at the Huntington Medical Research Institute for long-termstimulation in the peripheral and central nervous systems. Three importantfindings from this extensive work include the relationship between totalcharge and charge density, the comparison between capacitive electrodesand Faradaic electrodes, and the transient decreased neural excitability inresponse to electrical stimulation.

With respect to the amount of electrical charge that can be applied beforetissue damage is evident, there is a dependency between the charge perphase and the charge density. The charge per phase is the total chargeregardless of the electrode size.29 Charge density is obviously dependent onthe size of the electrode for a given amount of charge. When smaller electrodeareas are used, the amount of charge that can be used before neural damageoccurs corresponds to a charge density of 1 mC/cm2. However, with largerelectrodes, the neural damage threshold on charge limits the charge densityto 0.1 mC/cm2. It is hypothesized that some minimum amount of charge isrequired to result in damage to neural tissue. An analogy can be drawn tothe rheobase current, the minimum amount of current required to activatea neuron. If a very small amount of charge is used, the charge density canbe larger. This effect favors the use of smaller electrodes, because they cansupport a higher charge density with a smaller amount of charge.

The mechanism of neural damage is based on the type of electrode used.Briefly, faradaic electrodes use both capacitive coupling and reduction oxi-dation reactions to electrically stimulate tissue, while capacitive electrodesuse only capacitive current. A direct comparison of faradaic vs. capacitor



electrodes revealed that neural damage was evident with similar stimulusconditions for both electrodes. This suggests that the damage resulted fromthe passage of current through tissue, not from harmful bioproducts pro-duced by chemical reactions at the electrode.30

The last result of the Huntington group that will be discussed is that ofstimulus-induced decreased neural excitability.31 It was shown that, whenstimulating the cochlear nucleus continuously, the response threshold(recorded at the inferior colliculus) was increased with continuous pulsing.More current was required to elicit an equivalent response after several hoursof continuous pulsing. This decreased excitability could not be explainedhistologically. Also, the effect was transient; that is, with a night of rest theoriginal threshold was regained. This effect was only evident with continu-ous pulsing at a relatively high rate (more than 200 pulses/sec).

Electrochemical charge limits, the amount of charge that can be deliveredby the electrode prior to the formation of gas bubbles, are also empiricallydetermined by measuring the current vs. potential characteristics of theelectrode and by direct observation of the electrode surface during stimula-tion. This area has been studied extensively at EIC Labs in Newton, MA, byBrummer et al.,32–34 who have investigated a variety of materials for electricalstimulation. The material most commonly used for electrical stimulation isplatinum, which can safely supply 0.1 to 0.4 mC/cm2 of charge dependingon the pulse characteristics. Alternative materials that have higher limitssuch as iridium oxide (1 to 3 mC/cm2) and titanium nitride (0.6 to 0.9mC/cm2) have been studied extensively in simulated and real conditionsbut have not been widely used in commercial devices.

4.4.2 Selective stimulation of cells

Within the limits outlined above, it is possible to use electrical stimulationto effectively activate nerve cells, as demonstrated by cochlear implantsand deep brain stimulators. Short pulses of electrical current are used toactivate neurons and replace some functionality previously lost to disease.The shape of the current pulse can be manipulated to target specific cellsthat are not necessarily the cells closest to the electrode. This is becausedifferent types of cells have different response characteristics to electricalstimuli. In the retina, this can be readily demonstrated because differentcell types are grouped in well-organized layers. Greenberg35,36 demon-strated that using longer pulses allows bipolar cells to be selectively acti-vated over ganglion cells in the frog retina. This was determined by com-paring the strength duration curves from retinal stimulation under avariety of experimental conditions. The three experimental conditionstested were normal retina, light-saturated retina (photoreceptors hyperpo-larized), and retina with cadmium added (cadmium blocks synaptic trans-mission). In each case, a distinct strength–duration curve was obtained byrecording retinal ganglion cell responses (Figure 4.3). While all cells weredepolarized to some degree by electrical stimulation, the cell that initiated



the response was the cell for which depolarization resulted in action poten-tial generation in the ganglion cells.

In a normal retina, the retinal ganglion cell response could be generatedeither by direct electrical stimulation of the ganglion cells or by electricalstimulation of photoreceptor or bipolar cells and subsequent activation ofretinal ganglion cells through synaptic transmission. The presence of cad-mium strongly inhibits synaptic transmission, thus precluding a responseinitiated by any cell except a retinal ganglion cell. The strength–durationcurve generated with cadmium suggests that ganglion cells are more respon-sive at short pulse widths but quickly reach rheobase current. Light satura-tion of the retina hyperpolarizes the photoreceptor, making the bipolar cellmore likely to respond than the photoreceptor. However, because thestrength–duration curve for the light-bleached retina is different than thatfor the cadmium condition, the cell that initiates the response is likely to bedifferent. This argues that, in the absence of photoreceptor response, thebipolar cells will initiate a response according to the strength–duration curveshown. Comparing the cadmium strength–duration curve to the light-bleached strength–duration curve, we conclude that ganglion cells respondbetter to short-duration pulses and bipolar cells respond better to longerduration pulses. Put another way, bipolar cells have a lower rheobase, butganglion cells have a shorter chronaxie.

Other studies have examined the response of different cells to electricalstimulation. Weiland et al.37 showed that, in a human model of photoreceptordegeneration, electrical stimuation of a normal retina resulted in a different

Figure 4.3 Strength–duration curves obtained from electrical stimulation of the retinaunder various conditions. The retina was stimulated by an electrode on the ganglioncell surface, and action potentials were recorded from retinal ganglion cells in re-sponse to stimulation. The dark condition represents the state in which the retina iseasiest to excite with electrical stimulation. The light condition results in hyperpo-larization of the photoreceptors thereby decreasing the excitability of that part of theretina. The addition of cadmium blocks synaptic transmission, limiting the responsesite to the ganglion cells.



psychophysical perception than electrical stimulation of a laser-damagedretina. McIntyre and Grill12 have modeled the responses of spinal neuronsthat suggest a preference for soma stimulation with a cathodic first pulseand a preference for axonal stimulation with an anodic first pulse. Combined,these results suggest that, while proximity to the stimulating electrode is amajor factor in determining which cell is stimulated, it is not the only factorand it is possible to selectively activate types of neurons through manipu-lation of the stimulus pulse.

It is also possible to control the site of the electrical stimulus by usingmultipolar stimulation — that is, by using other electrode sites as the currentsink or current return, as opposed to a large distant electrode. Field theoryshows that a dipole electric field is more concentrated than a monopole field.However, in practice, bipolar stimulation has not been shown in cochlearimplants to improve the quality of the auditory sensation. In fact, somereports suggest that patients prefer monopolar stimulation.38 Cuff electrodes,devices that encircle a nerve fiber bundle with multiple electrodes aroundthe bundle, use current steering to target areas in the bundle. This is doneby passing electrical current between electrodes at specific points around thenerve, thereby activating nerve fibers between the two electrodes. Moresophisticated routines are possible that involve three to four electrodes toimprove selectivity.

4.5 Practical examples and applicationsIn practice, commercial electrical stimulation systems implement some ofthe experimental and theoretical findings discussed above. Cochlearimplants use charge-balanced, biphasic stimulation pulses to activate theauditory nerve. One important result from cochlear implant research wasthe effectiveness of pulses of short time duration. Early cochlear implantsproduced an electrical sinusoid at the same frequency as the sound wave itwas trying to replicate. If a sound wave at 200 Hz was present, then thecochlear implant stimulated with a 200-Hz sine wave. It was shown, how-ever, that short stimulus pulses of 0.1 msec presented at 200 pulses/sec werejust as effective, in most cases, in replicating the sensation of a 200-Hzsound.39–41 The advantage of using a 0.1-msec pulse vs. a 200-Hz (or 50-msec)sinusoid is the savings in electrical charge used. Because it was also notedthat in some cases the sinusoid was more effective, some cochlear implantshave the ability to stimulate with both pulse and sinusoid stimuli.

Another example of successful clinical application of neural stimula-tion is the deep brain stimulator (DBS). This device is currently beingevaluated for the treatment of Parkinsonian tremor.42 Seizures in the tha-lamic region of the brain are believed to cause some types of tremor thatcan render the afflicted individual unable to control movements. By elec-trically stimulating the thalamus, it is hypothesized that the seizures aremoderated or depressed, allowing other areas of the brain to functionproperly to allow voluntary motor control. While the exact mechanism of



DBS is not clear, the results can be dramatic. In a matter of minutes,Parkinson’s patient can go from being unable to stand due to tremor tobeing able to easily walk across a room.

Both the cochlear implant and deep brain stimulator demonstrate whatshould be a guiding principle for neural prosthetic devices. The purposeof the device is not to closely replicate the tissue it is replacing; rather, thepurpose of the device is to replicate the lost function, regardless of howthis is accomplished. We do not know how to build a cochlea, a thalamus,or a retina, but we do have some knowledge of how to electrically activatenerves. Using the established boundaries of safe stimulation, it is possibleto achieve dramatic results in patients who otherwise would have no hopeof a treatment.

References1. Djourno, A. and Eyries, C., Prothese auditive par excitation electrique a dis-

tance du nerf sensorial a l’aide d’un bobinage inclus a demeure, Presse Med.,35, 14–17, 1957.

2. Brindley, G. and Rushton, D., Implanted stimulators of the visual cortex asvisual prosthetic devices, Trans. Am. Acad. Ophthalmol. Otolaryngol., 78,OP741–OP745, 1974.

3. House, W.F., Cochlear implants, Ann. Otol. Rhinol. Laryngol., 85(suppl. 27, pt.2), 1–93, 1976.

4. Dobelle, W.H., Artificial vision for the blind by connecting a television camerato the visual cortex, Am. Soc. Artificial Internal Organs J., 46, 3–9, 2000.

5. Dobelle, W.H., Quest, D.O., Antunes, J.L., Roberts, T.S., and Girvin, J.P., Ar-tificial vision for the blind by electrical stimulation of the visual cortex, Neu-rosurgery, 5(4), 521–527, 1979.

6. Plonsey, R. and Barr, R.C., Bioelectricity, A Quantitative Approach, Plenum Press,New York, 1991.

7. Kandel, E.R., Schwartz, J.H., and Jessell, T.M., Principles of Neural Science, 3rded., Elsevier Science, New York, 1991.

8. Hodgkin, A. and Huxley, A., A quantitative description of membrane currentand its application to conduction and excitation in nerve, J. Physiol., 117,500–544, 1952.

9. van Den, H.C. and Mortimer, J.T., Generation of unidirectionally propagatedaction potentials in a peripheral nerve by brief stimuli, Science, 206(4424),1311–1312, 1979.

10. Reilly, J.P., Sensory responses to electrical stimulation, in Reilly, J.P., Ed.,Applied Bioelectricity: From Electrical Stimulation to Electropathology, Springer-Verlag, New York, 1998, pp. 240–298.

11. McNeal, D.R., Analysis of a model for excitation of myelinated nerve, IEEETrans. Biomed. Eng., BME-23(4), 329–336, 1976.

12. McIntyre, C.C. and Grill, W.M., Selective microstimulation of central nervoussystem neurons, Ann. Biomed. Eng., 28(3), 219–233, 2000.

13. Neher, E., Sakmann, B., and Steinbach, J.H., The extracellular patch clamp, amethod for resolving currents through individual open channels in biologicalmembranes, Pflügers Arch., 375(2), 219–228, 1978.



14. Neher, E. and Sakmann, B., Single-channel currents recorded from membraneof denervated frog muscle fibres, Nature, 260(5554), 799–802, 1976.

15. Marshall, S. and Skitek, G., Electromagnetic concepts and applications, 2nded., Prentice-Hall, Englewood Cliffs, NJ, 1987.

16. Burton, M.J., Miller, J.M., and Kileny, P.R., Middle-latency responses. I. Elec-trical and acoustic excitation, Arch. Otolaryngol. Head Neck Surg., 115(1), 59–62,1989.

17. Neuman, M.R., Biopotential amplifiers, in Medical Instrumentation: Applicationand Design, Webster, J.G., Ed., Houghton-Mifflin, Boston, 1992, pp. 288–353.

18. Parsa, V., Parker, P.A., and Scott, R.N., Adaptive stimulus artifact reductionin noncortical somatosensory evoked potential studies, IEEE Trans. Biomed.Eng., 45(2), 165–179, 1998.

19. Takashima, I., Kajiwara, R., and Iijima, T., Voltage-sensitive dye versus intrin-sic signal optical imaging: comparison of optically determined functionalmaps from rat barrel cortex, NeuroReport, 12(13), 2889–2894, 2001.

20. Tominaga, T., Tominaga, Y., and Ichikawa, M., Simultaneous multi-site re-cordings of neural activity with an inline multi-electrode array and opticalmeasurement in rat hippocampal slices, Pflügers Arch., 443(2), 317–322, 2001.

21. Kohn, A., Metz, C., Quibrera, M., Tommerdahl, M.A., and Whitsel, B.L.,Functional neocortical microcircuitry demonstrated with intrinsic signal op-tical imaging in vitro, Neuroscience, 95(1), 51–62, 2000.

22. Dragunow, M. and Faull, R., The use of c-fos as a metabolic marker in neuronalpathway tracing, J. Neurosci. Meth., 29(3), 261–265, 1989.

23. Yoshida, K., Imaki, J., Fujisawa, H., Harada, T., Ohki, K., Matsuda, H. et al.,Differential distribution of CaM kinases and induction of c-fos expression byflashing and sustained light in rat retinal cells, Invest. Ophthalmol. Vis. Sci.,37(1), 174–179, 1996.

24. Saito, H., Miller, J.M., Pfingst, B.E., and Altschuler, R.A., Fos-like immunore-activity in the auditory brainstem evoked by bipolar intracochlear electricalstimulation: effects of current level and pulse duration, Neuroscience, 91(1),139–161, 1999.

25. Saito, H., Miller, J.M., and Altschuler, R.A., Cochleotopic fos immunoreactivityin cochlea and cochlear nuclei evoked by bipolar cochlear electrical stimula-tion, Hearing Res., 145(1–2), 37–51, 2000.

26. West, D.C. and Wolstencroft, J.H., Strength-duration characteristics of myeli-nated and non-myelinated bulbospinal axons in the cat spinal cord, J. Physiol.(London), 337, 37–50, 1983.

27. Rubinstein, J.T., Spelman, F.A., Soma, M., and Suesserman, M.F., Currentdensity profiles of surface mounted and recessed electrodes for neural pros-theses, IEEE Trans. Biomed. Eng., BME-34(11), 864–875, 1987.

28. Lilly, J.C., Injury and excitation by electric currents: the balanced pulse-pairwaveform, in Electrical Stimulation of the Brain, Sheer, D.E., Ed., Hogg Foun-dation for Mental Health, 1961.

29. McCreery, D.B., Agnew, W.F., Yuen, T.G.H., and Bullara, L., Charge densityand charge per phase as cofactors in neural injury induced by electricalstimulation, IEEE Trans. Biomed. Eng., 37(10), 996–1001, 1990.

30. McCreery, D.B., Agnew, W.F., Yuen, T.G., and Bullara, L.A., Comparison ofneural damage induced by electrical stimulation with faradaic and capacitorelectrodes, Ann. Biomed. Eng., 16(5), 463–481, 1988.



31. McCreery, D.B., Yuen, T.G., Agnew, W.F., and Bullara, L.A., A characterizationof the effects on neuronal excitability due to prolonged microstimulation withchronically implanted microelectrodes, IEEE Trans. Biomed. Eng., 44(10),931–939, 1997.

32. Brummer, S.B., Robblee, L.S., and Hambrecht, F.T., Criteria for selecting elec-trodes for electrical stimulation, theoretical and practical considerations, Ann.N.Y. Acad. Sci., 405, 159–171, 1983.

33. Robblee, L.S., Lefko, J.L., and Brummer, S.B., Activated Ir: an electrode suit-able for reversible charge injection in saline solution, J. Electrochem. Soc.,130(3), 731–732, 1983.

34. Brummer, S.B. and Turner, M.J., Electrochemical considerations for safe elec-trical stimulation of the nervous system with platinum electrodes, IEEE Trans.Biomed. Eng., 24(1), 59–63, 1977.

35. Greenberg, R.J., Analysis of electrical stimulation of the vertebrate retina:work towards a retinal prosthesis, The Johns Hopkins University, Baltimore,MD, 1998.

36. Greenberg, R.J., Velte, T.J., Humayun, M.S., Scarlatis, G.N., de JE, Jr., A com-putational model of electrical stimulation of the retinal ganglion cell, IEEETrans. Biomed. Eng., 46(5), 505–514, 1999.

37. Weiland, J.D., Humayun, M.S., Dagnelie, G., de JE, Jr., Greenberg, R.J., Iliff,N.T., Understanding the origin of visual percepts elicited by electrical stim-ulation of the human retina, Graefes Arch. Clin. Exp. Ophthalmol., 237(12),1007–1013, 1999.

38. Pfingst, B.E., Franck, K.H., Xu, L., Bauer, E.M., and Zwolan, T.A., Effects ofelectrode configuration and place of stimulation on speech perception withcochlear prostheses, J. Assoc. Res. Otolaryngol., 2(2), 87–103, 2001.

39. Wilson, B.S., Finley, C.C., Lawson, D.T., and Zerbi, M., Temporal representa-tions with cochlear implants, Am. J. Otol., 1997. 18(6 Suppl), S30-S34.

40. Wilson, B.S., Finley, C.C., Lawson, D.T., Wolford, R.D., and Zerbi, M., Designand evaluation of a continuous interleaved sampling (CIS) processing strat-egy for multichannel cochlear implants, J. Rehabil. Res. Dev., 30(1), 110–116,1993.

41. Wilson, B.S., Finley, C.C., Farmer, J.C., Jr., Lawson, D.T., Weber, B.A., Wolford,R.D. et al., Comparative studies of speech processing strategies for cochlearimplants, Laryngoscope, 98(10), 1069–1077, 1988.

42. Koller, W., Pahwa, R., Busenbark, K., Hubble, J., Wilkinson, S., Lang, A. et al.,High-frequency unilateral thalamic stimulation in the treatment of essentialand parkinsonian tremor, Ann. Neurol., 42(3), 292–299, 1997.



chapter five

Extracellularelectrical stimulation of central neurons: quantitative studies

Dongchul C. Lee, Cameron C. McIntyre, and Warren M. Grill

Contents

5.1 Introduction5.1.1 What elements are stimulated in the CNS?5.1.2 Selective stimulation of CNS elements5.1.3 Computational modeling as a tool for understanding

and design5.2 Quantitative models of CNS neurons

5.2.1 Physical basis of models: from cells to circuits5.2.2 Geometric properties of a range of CNS neurons5.2.3 Cable models of CNS neurons5.2.4 Cable equation: continuous form5.2.5 Cable equation: discrete form5.2.6 Assumptions5.2.7 Passive electrical properties of CNS neurons5.2.8 Intracellular vs. extracellular stimulation5.2.9 Source driving polarization5.2.10 Modeling of synaptic inputs

5.3 Properties of CNS stimulation5.3.1 Direct activation5.3.2 Indirect effects

5.4 Selective stimulation5.4.1 Effect of stimulus frequency



96 Handbook of Neuroprosthetic Methods

5.4.2 Effect of stimulus waveform5.5 ConclusionAcknowledgmentsReferences

5.1 IntroductionElectrical activation of the nervous system is a method to restore functionto persons with neurological disorders due to disease or injury and a tech-nique to study the form and function of the nervous system. Application ofelectrical stimulation, in the form of neural prostheses, is used to restoreboth motor and sensory functions. Although restoration of motor functionhas primarily been accomplished by activation of peripheral motor nervefibers (i.e., last-order neurons), to restore complex motor functions it may beadvantageous to access the nervous system at a higher level and use theintact neural circuitry to control the individual elements of the motor sys-tem,22 which may be accomplished by intraspinal microstimulation.17 Simi-larly, restoration of the sense of vision may be accomplished by intracorticalstimulation in the visual cortex,3,55 and restoration of the sense of hearing bystimulation of the ventral cochlear nucleus.30

In applications of central nervous system (CNS) stimulation, electrodesare positioned in geometrically and electrically complex volume conductors(the brain and spinal cord) containing cell bodies, dendrites, and axons inclose proximity. When a stimulus is applied within the CNS, cells and fibersover an unknown volume of tissue are activated, resulting in direct excitationas well as trans-synaptic excitation/inhibition from stimulation of presyn-aptic axons and cell bodies. Fundamental knowledge of the interactionsbetween the applied currents and the neurons of the CNS has been limited,and this void will impair the development of safe and effective futuredevices. In this chapter we review the issues that arise during stimulationof the CNS and use quantitative computational modeling to establish afoundation upon which to build future applications of electrical stimulationin the brain. We will focus on two fundamental questions that arise duringCNS stimulation:

1. What elements are stimulated by extracellular electrodes in the CNS?2. What methods will enable selective stimulation of different neuronal

elements in the CNS?

5.1.1 What elements are stimulated in the CNS?

The question of what elements are stimulated by extracellular electrodes inthe CNS was addressed in the seminal review of Ranck,45 and we revisit andexpand upon the findings set forth in that review. The neuronal elements ofinterest (Figure 5.1) include local cells around the electrode, including those



projecting locally around the electrode as well as those projecting away fromthe region of stimulation, axons passing by the electrode, and presynapticterminals projecting onto neurons in the region of the electrode. Effects ofstimulation can be mediated by activation of any or all of these elementsand include direct effects of stimulation of postsynaptic elements, as well asindirect effects mediated by electrical stimulation of presynaptic terminalsthat mediate the effects of stimulation via synaptic transmission.

Previous experimental evidence demonstrates that different neuronalelements have similar thresholds for extracellular stimulation and illus-trates the need for the design of methods that would enable selectivestimulation. In vitro measurements in cortical brain slices indicate that cellsand fibers have similar thresholds for activation.36,37 Similarly, in vivo mea-surements using microstimulation indicate that fibers and cells have similarthresholds for cathodic rectangular stimuli.23,45,51 Recent computationalstudies of the excitation of CNS neurons indicate that, with conventionalrectangular stimuli, axons of passage and local cells respond at similarthresholds.32,33 Also, the thresholds for generating direct and synaptic exci-tation of neurons in the spinal cord,23 red nucleus,4 and cortex26 were quitesimilar. Extracellular activation of type-identified spinal motoneurons indi-

Figure 5.1 During electrical stimulation of the central nervous system, electrodes arepositioned within heterogeneous populations of neuronal elements. Neuronal ele-ments that can be activated by stimulation include local cells around the electrodeprojecting away from the region of stimulation (A), local cells around the electrodeprojecting locally (B), axons passing by the electrode (C), and presynpatic terminalsprojecting onto neurons in the region of the electrode (D).



cated that the current amplitude necessary to induce repetitive firing witha monopolar electrode was not significantly different for type-S and type-F motoneurons,58 and modeling results demonstrated a very small differ-ence between the extracellular thresholds of different sized neurons.33 Thus,with conventional stimulation techniques, the thresholds for activation ofdifferent neuronal elements are quite similar, and it is difficult to isolatestimulation of particular neurons.

5.1.2 Selective stimulation of CNS elements

A neural prosthesis using microstimulation of the CNS will require selec-tive and controlled activation of specific neural populations. The “com-plexity of the spinal circuitry implies that if a supraspinal trigger couldgenerate movement it would need to be a highly focused drive onto aselect populations of interneurons.”7 Similarly, interpretation of physio-logical investigations employing microstimulation requires knowledge ofthe effects of stimulation on different neuronal elements. Application ofextracellular currents may activate or inactivate (block) neurons and oraxons dependent on their morphology, distance from the electrode, ori-entation with respect to the electrode, and discharge rate, as well as thestimulus parameters.45 Effects on cells may differ from the effects on fibersof passage, and fiber activation will result in both antidromic and ortho-dromic propagation. While the technology for fabrication of high-densityarrays of microelectrodes for insertion in the CNS has advanced greatly,2,9

our knowledge of neuronal activation patterns has not. Advancing ourunderstanding of neuronal excitation and determining what is requiredto target excitation we will provide useful design parameters for micro-electrode arrays.

5.1.3 Computational modeling as a tool for understanding and design

Computational modeling provides a powerful tool to study extracellularexcitation of CNS neurons. The volume of tissue stimulated, both for fibersand cells, and how this volume changes with electrode geometry, stimulusparameters, and the geometry of the neuronal elements are quite challengingto determine experimentally. Using a computer model enables examinationof these parameters under controlled conditions. Neural modeling providesa powerful tool to address the effects of stimulation on all the different neuralelements around the electrode simultaneously. Further, well-designed mod-eling studies enable generation of experimentally testable hypothesesregarding the effects of stimulation conditions on the pattern and selectivityof neuronal stimulation within the CNS. However, the strengths of modelingare tempered by the necessary simplifications made in any reasonable model.In turn, modeling should be coupled as closely as possible to experimentalwork enabling a synergistic analysis of results.



The utility of such an approach has been demonstrated by previous field-neuron models of cochlear stimulation which were able to replicate experi-mental activation patterns and document the effects of changes in electrodegeometry and stimulus parameters.14–16 Similarly, integrated field-neuronmodels of epidural spinal cord stimulation have been used to explain clinicalpatterns of paresthesias10,11,57,59 and to design novel electrode geometries forselective stimulation of targeted neural elements.25,60 These examples dem-onstrate the utility of computer-based modeling in understanding and con-trolling neural elements activated by electrical stimulation.

5.2 Quantitative models of CNS neurons5.2.1 Physical basis of models: from cells to circuits

Neural cells in the CNS are highly varied in their form and function but allshare the properties of excitability and polarization. Excitability is based onselective ion channels and polarization is based on the concentration differ-ences of ions, generated by ionic pumps, across the cell membrane. Accord-ing to Robertson’s proposition,52 all membranes or major portions of mem-branes have a common basic structure. This structure includes a lipid bilayercovered by non-lipid monolayers on both sides. Within the lipid bilayerreside proteins, and transmembrane proteins form the pores that enable ioniccurrent across the membrane. The electrical properties of the membrane aresimilar to those of a parallel RC circuit, the so-called passive membrane, wherethe capacitor represents the lipid bilayer and the resistor represents thetransmembrane ion channels.

The excitability of neurons comes from the ion channels. Ion channelscontribute to changing the conductance across the membrane by passingspecific ions. The conductance of some ion channels is dependent on thevoltage across the membrane and thus can be considered as nonlinear resis-tors.24 The nonlinear properties of ion channels, such as the sodium channel,can produce regenerative coupling to the transmembrane potential by thegreatly increased conductance to sodium ions. This nonlinear property inconductance of specific ions is referred to as active membrane.

The passive and active properties are the basic components of the mem-brane throughout the neuron including the cell body, dendrites, and axon.The dendrite and axon are long cylindrical tubes filled with cytoplasm,which has a higher electrical conductivity than the extracellular fluid. Also,for axons and dendrites the electric current flowing through the membraneis much less than the current flowing parallel to the cylinder axis becausethe resistance of the membrane is much higher than the cytoplasmic resis-tance. Therefore, the nerve cell with passive membrane properties (i.e., notconsidering nonlinear ion channels) can be considered a good conductorinsulated by a membrane that has high resistivity and a certain capacity.This analog bridges the investigation of electrophysiological properties ofneurons to cable theory (Figure 5.2).



5.2.2 Geometric properties of a range of CNS neurons

The types of neurons in the CNS are exceedingly diverse and are classifiedinto three large groups by shape as unipolar, bipolar, and multipolar neurons.Each group has common features of dendrite and axon structure with respectto the cell body. The morphology of neurons has an impact on their responseto extracellular stimulation because the entire neuron is exposed to theelectric field. Therefore, the response of the cell is modulated by influencesfrom every branch exposed to the electric field. Table 5.1 shows examples ofmammalian CNS neurons to indicate the range of cell structure and size.

5.2.3 Cable models of CNS neurons

Cable theory was first presented in 1855 by William Thomson, who providedthe mathematical derivation and applications for submarine telegraphiccables. This theory included both steady-state and transient solutions for par-ticular boundary conditions and initial conditions.27,41 The solution was for asingle spatial dimension that facilitated the theoretical treatment of transientas well as steady-state solutions. The theory was applied for nerve electronusby Matteucci in 1863 and further developed by Hermann in the 1870s.

5.2.4 Cable equation: continuous form

The continuous form of the cable equation is derived from a compartmentmodel that consists of a series of compartments, each with a resistance anda capacitance.41 Each compartment represents a segment of cable with thelength of

∆x, and all properties such as resistance and capacitance in thesegment are lumped into one element for each (Figure 5.3). The cable equa-tion is derived by application of Kirchoff’s current law (KCL) — conservationof currents — which states that the difference in the currents (between ii1

and ii2) in the axial direction is the current flowing through the membrane(Figure 5.4). Mathematically this is expressed as:

ii1 – ii2 = –

∆ii = im

∆x

The membrane current (im

∆x) is the sum of two components: the resistivecurrent Vm(

∆x/rm) and the capacitive current cm

∆x((

∂Vm)/(

∂t)). Therefore, themembrane current is given by:

Figure 5.2 Typical neuron and cable model. The passive electrical properties of thedendrites, soma, and axon are similar to the core cable with insulation.



Table 5.1 Range of Geometrical Properties of CNS Neurons

Region Neuron Ref.Mean Soma

Diameter (µm)

Numberof Dendrite

Stems on Soma

DendriticTerminal

Length (µm)

Cerebellum Purkinje cell Roth and Hauser54 50 ~ 80 2 ~ 10 50 ~ 200Thalamus TCP neurons Ohara and Havton38 11 ~ 20 4 ~ 8 28 ~ 3000Hippocampus Pyramidal cell Bannister and Larkman;5 Bilkey and

Schwartzkroin615 ~ 30 2 ~ 8 300 ~ 1000

Retina Ganglion cell Sheasby and Fohlmeister56 20 ~ 30 3 ~ 7 50 ~ 1000Spinal cord Motor neuron Rose et al.53 50 ± 10 10 ± 2 1,150 ± 304

1100_Book Page 101 T



The equation is simplified by dividing

∆x on both sides and the result is:

The axial current, derived from Ohm’s law as that current flowing betweentwo consecutive nodes (Vi1 and Vi2 or Vi2 and Vi3), is defined by potentialdifference divided by resistance and expressed as:

Figure 5.3 Compartment model of cable. Insulation (membrane) around core isequivalent to resistance in parallel with capacitance.

Figure 5.4 Application of Kirchoff’s current law at a compartment of a cable modelto derive the cable equation. The membrane current includes capacitive and resistivecurrents, and their sum is equal to the change in the axial current.

i x Vx

rc x

Vtm m

mm

m∆ ∆ ∆=

+

∂∂

iVr

cVtm

m

mm

m= +∂∂

V V V i r x

i rVx

i i i i i

i ii

1 2− = − =

= −

∆ ∆

∆∆



By taking the limit

∆x

→ 0, the axial current is expressed by partial derivative:

Going back to KCL, defining conservation of the current, the membranecurrent with partial derivative form yields:

Therefore, the membrane current is the partial derivative of the axial currentwith respect to x. Again, the axial current is also a partial derivative of the axialpotential distribution by spatial variable x. Applying the axial current equationin partial derivative form (–(

∂ii/

∂x)),the membrane current can be expressed as:

The transmembrane potential (Vm) is defined as Vi – Ve, so the intracel-lular potential Vi can be replaced by Vm + Ve. Finally, the current im isexpressed as:

Combining with the earlier equation for transmembrane current as thesum of the resistive and capacitive current yields:

The general single-dimension cable equation is a partial differentialequation expressed as:

ir

Vxi

i

i= −∂∂

1

iix

ix

mi

i

= −

= −∂∂

→

∆∆

∆, when x 0

iix x

Vr x

Vr xm

i i

i

i

i

= −∂∂

= − ∂∂

−∂∂

=

∂∂

2

2

iV

r xV

r xmm

i

e

i

=∂

∂+

∂∂

2

2

2

2

1 1

1 1

2

2

2

2

2

2

2

2

rVx r

Vx

Vr

cVt

rVx

Vr

cVt r

Vx

i

m

i

e m

mm

m

i

m m

mm

m

i

e

∂∂

+∂∂

= +∂∂

∂∂

− −∂∂

= −∂∂

λ τ22

2

∂∂

− − ∂∂

=Vx

VVt

F

1100_ch05.fm Page 103 Tuesday, November 5, 2002 8:06 AM


where V is the transmembrane voltage as a function of x (spatial variable)and t (time), and λ = and τ = rmcm are the space and time constantsdefined by the electrical properties of the neuron. The F represents a forcingfunction or input caused by synaptic conductance change, applied electricfields, and/or active membrane properties. This nonhomogeneous partialdifferential equation can be transformed to a homogeneous equation (F = 0)with additional initial conditions by the principle of superposition. Furthersimplification can be made for steady-state conditions by (∂V/∂t) = 0, whichwill yield a homogeneous linear second-order ordinary differential equationas λ2(∂2V/∂x2) – V = 0, the solution of which is a sum of two exponentials(cable of infinite length) or hyperbolic functions (cable of finite length). Thesolution for a finite cable is:

where B1 and B2 are defined by boundary condition at x = 0 and x = l (cablelength). The equivalent solution for the infinite cable is:

where A1 and A2 are determined by conditions at x = 0 and x = ±∞. Transientsolutions of the cable equation can be obtained by using separation of vari-ables. For a finite-length cable with the boundary condition of sealedends,40,43,50 the solution can be expressed as an infinite series:

where l and τ are the length of cable and time constant of cable, respectively.

5.2.5 Cable equation: discrete form

The continuous form of the cable equation is limited to homogeneous struc-tures with restrictive assumptions. The passive properties of the dendritictree and unmyelinated axons with constant diameter can be modeled using

rm ri⁄

V Bx

Bx= +1 2cosh( ) sinh( )

λ λ

V Ax

Ax= + −1 2exp( ) exp( )

λ λ

V B nx t n

lt

Bl

V x t dx

Bnl

V x tn x

ldx n

nn

l

l

= − −

= =

= = >

=

∞

∑

∫

∫

cos( )exp( ( ) )

( ) ( , )

( ) ( , )cos( ) ,

πλ τ

πλτ

π

0

2

0

0

0

10

20 0



the continuous cable equation. However, a typical dendritic tree has variousdiameters from stem to dendritic terminal, and myelinated axons have dif-ferent membrane properties at the nodes of Ranvier and in the internodalsegments. The nonhomogenous cable properties can be modeled using adiscrete form of the cable equation that is a mathematical representation ofthe compartmental circuit model of a neuron.

The general discrete cable equation is directly derived from the continuouscase. The partial derivative of the spatial variable is replaced by ∆, yielding:

This equation is only useful for compartmental models of homogeneouscables, but small modifications of the equation give rise to great flexibilityfor practical applications such as nonhomogeneous dendrites, cell bodies,and myelinated axons. In the discrete compartmental cable model, the resis-tance (axial and membrane) and capacitance are not required to be functionsof ∆x. Therefore, each variable — 1/(ri∆x), cm∆x, and ∆x/rm — can be replacedby Ga(k∆x) (Ω∠1), Cm(k∆x) (F), and Rm(k∆x) (Ω). This yields the discrete cableequation applicable to a cable with any geometry. The steady-state solutionof the discrete cable equation can be obtained from linear algebra; however,the transient solution must be obtained numerically.

As an example of application of the discrete cable equation, a myelinatedaxon is considered where the myelin has a very high resistivity (assumed tobe an insulator), thus simplifying the model to include only the nodes ofRanvier and the axial resistance (Figure 5.5). Under this circumstance, thecable equation in differential form will be discrete in space and continuousin time. The membrane (nodal) current is defined similarly to the continuouscase with a resistive current (GmVm) and a capacitive current (Cm(dVm/dt)).The total current flowing through the node of Ranvier is expressed as:

Figure 5.5 Compartmental model of a myelinated axon. The internodal segment ismodeled by an axial resistor, because of the highly resistive myelin sheath. The nodeof Ranvier is lumped into a single compartment, which is a relatively small portioncompared to the internodal segment (1.5 µm in length vs. 1 mm in length).

1 12

2

2

2rVx

Vr

cdVdt r

Vxi

m m

mm

m

i

e∆∆

∆∆

− −

= −

I G V CdVdtnode m m m

m= +



The axial current between successive nodes of Ranvier depends on the inter-nodal conductance (Ga) and potential difference (Vi1 – Vi2 or Vi2 – Vi3) accord-ing to Ohm’s law. Just as the membrane current was obtained from thedifference of axial currents in the continuous case, the nodal current equationin the discrete model is:

where Vm = Vi – Ve. Combining this equation with the equation for thecomponents of the membrane current yields:

Rearranging terms and defining the second difference of the potential as:

yields the discrete cable equation for myelinated axon:

The final equation is similar to the general form of the cable equationwith a forcing term of Ga ⋅ ∆2Ve. For a step change in the extracellular potentialfrom steady state, as would occur by application of a rectangular currentpulse, the forcing term will determine initial polarization pattern of the axon,because the Cm(dVm/dt) term will follow the sign of the forcing term.

5.2.6 Assumptions

The cable model and cable equation are good approximations to under-stand the properties of neurons, but they must be used with the followingassumptions:

1. Any angular or radial dependence of V within the core or outsidethe membrane is neglected. This assumption is reasonable for steady-state solutions of small-diameter cables with high-conductivity corescorresponding to high cytoplasm conductivities. Low conductivityof the cable core will generate a voltage gradient within the cable in

I I I Ga V V Ga V V

Ga V V V

Ga V V V Ga V V V

node i i i i i

i i i

m m m e e e

= − = ⋅ − − ⋅ −

= ⋅ − +

= ⋅ − + + ⋅ − +

2 1 2 2 3

1 2 3

1 2 3 1 2 3

2

2 2

( ) ( )

( )

( ) ( )

G V CdVdt

Ga V V V Ga V V Vm m mm

m m m e e e+

= ⋅ − + + ⋅ − +( ) ( )1 2 3 1 2 32 2

Ga V V V Ga Vm m m m⋅ − + = ⋅( )1 2 322 ∆

G V CdVdt

Ga V Ga Vm m mm

m e+

− ⋅ = ⋅( )∆ ∆2 2



the radial direction and violate this assumption. Practically, the highconductivity of the cytoplasm eliminates this effect. This assumptionis also invalid under extracellular stimulation with large diametercables such as used to represent the soma. The extracellular potentialsare dependent on the angular position on the soma,28,29 and the an-gular dependence of V is a function of the electrode to cell distanceand of the electrode type. Therefore, it is valid only for relativelysmall diameter cables such as dendrites and axons or cables far fromthe stimulation electrode.

2. The core conductivity is uniform everywhere (homogeneous medi-um). This assumption implies that the ratio between the resistivityof cytoplasm and the cross sectional area is constant. For homoge-neous cytoplasm, it implies constant diameter of the cable. Practically,the diameters of dendrites and axons may change along their paths.Thus, the morphology of neurons limits direct application of analyt-ical solution of the continuous cable equation and requires discreti-zation to account for geometric inhomogeneity.

5.2.7 Passive electrical properties of CNS neurons

In the nervous system, information is transported long distances by actionpotentials. The action potential is generated by depolarization of the mem-brane where active ion channels are present. The depolarization can begenerated by any input, including synaptic current or current generated byan extracellular electric field. When the degree of polarization from a certaininput (voltage or current) is under threshold the neuron will not fire an actionpotential and can be considered using the so-called passive electrical properties.The passive electrical properties are described by the input impedance, thetime constant, and the resting potential. The input impedance of the neuronis defined by the relationship between the current applied by an intracellularelectrode and the transmembrane voltage response. The input impedancedetermines how much the transmembrane voltage of neuron will change inresponse to a steady current:

As the passive membrane is modeled as a resistance in parallel with acapacitance, the response of the membrane (voltage) takes time to reachsteady state. The change in transmembrane potential of the passive mem-brane can be described by following equation:

where τ is the membrane time constant given by the product of input resis-tance and input capacitance. The resting potential is determined by the ionic

∆V I Rin= ×

∆V t I R einjected in

t( ) ( )= − −1 τ



Table 5.2 Range of Passive Electrical Properties of CNS Neurons

Region Neuron Ref.Input Resistance

(MΩ)Time Constant

(msec)Resting Potential

(mV)

Cerebellum Purkinje cell Raman and Bean;44 Rothand Hausser54

59.3 ± 38.4 64.6 ± 17.2 –62 ± 3 with 300 nM TTX

Thalamus Thalamocortical neuron

Turner et al.61 30 ~ 240 5 ~ 35 –60 ~ –73

Hippocampus Pyramidal cell Bilkey and Schwartzkroin6

23 ± 2 12.5 ± 0.97 –55.7 ± 1.28

Retina Ganglion cell Sheasby and Fohlmeister56

800 ~ 1600 47 ~ 85 –64 ~ –66

Spinal cord Motor neuron McDonagh et al.31 2.5 ~ 32 2.5 ~ 52 –60 ~ –83




concentration difference across the membrane and the states of the ion chan-nels at rest. The balance of ion fluxes gives rise to the resting potential, whichis quantified by the Goldman equation. It usually ranges from –60 to –70mV. The resting potential is a reference point for measurements of changesin transmembrane potential. The range of passive electrical properties ofexample neurons presented in Table 5.2.

5.2.8 Intracellular vs. extracellular stimulation

Properties of neuronal excitation have been studied primarily by intra-cellular electrical stimulation by injecting current through a glass pipetteelectrode. The injected current will flow in the axoplasm and pass throughmembrane, and the neuron can again be modeled using a discrete compart-mental cable model (Figure 5.6). The transmembrane voltage at each com-partment is determined by current flowing through the membrane and axo-plasm, and an analytical solution provides the profile of transmembranevoltage along the cable for different boundary conditions (Figure 5.7).

Figure 5.6 Electrically equivalent model to intracellular stimulation. Intracellularstimulation of the nerve cell or fiber is modeled as a cable with a current or voltagesource connected to specific locations of the cable. The stimulation corresponds tothe forcing term (F) in the cable equation.

Figure 5.7 Analytical solution of the cable equation with three different boundaryconditions to determine the profile of transmembrane potential generated by intra-cellular current injection.



Application of current in the extracellular space (extracellular stimu-lation) creates a specific pattern of electric field and thus electric potentialsalong the cable (Figure 5.8). The extracellular potentials (Ve) along the cableare determined by the geometry of the cable, the source geometry andlocation, and the electrical properties of the extracellular space. As derivedin the cable equation,1,41,48 the second derivative of the extracellular poten-tials along the cable (–(1/ri) (∂2Ve/∂x2)) will produce current across themembrane and intracellular axial current and thus changes in transmem-brane potential.

Mathematically this forcing term (activating function: (1/ri) (∂2Ve/∂x2) isequivalent to current injection in each compartment in an equivalent modelwith intracellular current injections.48,62 As we derived the continuous anddiscrete cable equations using Kirchoff’s current law, the conservation ofcurrent at each node must be satisfied. Under the condition that the secondderivative of Ve equals zero, the cable equation is homogeneous and consistsof three terms:

The first term represents the axial current difference, the second representsthe membrane resistive current, and the third represents the membranecapacitive current with unit of A/cm.2 Therefore, the additional term, –(1/ri)(∂2Ve/∂x2) arising from the extracellular potentials corresponds to a current(A/cm2) source connected to each node, as shown in Figure 5.9. If a seriesof equivalent currents are injected along the cable, the effect of extracellularstimulation is equivalent to the effect of intracellular stimulation with a seriesof distributed intracellular current sources. The equivalent current at eachnode can be of anodic or cathodic polarity from a single extracellular elec-trode, because the forcing term (Ga ⋅ ∆2Ve) is determined by axon geometryand electrode type and location.62

The potentials generated by an extracellular electrode produce bothinward and outward transmembrane current in the cable, as shown in Figure5.10. The direction of transmembrane current depends on the location alongthe cable and creates regions of both depolarization and hyperpolarization

Figure 5.8 Electrically equivalent cable model under extracellular stimulation. Theextracellular electrode will determine the extracellular potentials (Ve) along the cablewhich act as the sources to generate axial and transmembrane currents.

10

2

2rVx

Vr

CVti

m m

mm

m∂∂

− −∂∂

=



along the cable, in contrast to the unidirectional polarization resulting fromintracellular stimulation (Figure 5.7).

5.2.9 Source driving polarization

Solution of the infinite cable equation provides the steady-state voltagedistribution along the cable expressed as an exponential function. The changein voltage along the cable is determined by the space constant, which is afunction of the ratio between the membrane and cytoplasmic resistivities(Figure 5.7). For realistic dendrites (finite cable with nonhomogeneous geo-

Figure 5.9 Equivalent model of extracellular stimulation with current injection ineach compartment. (A) The equivalent current for each compartment is derived fromthe cable equation with the non-zero Ve using Kirchoff’s current law. (B) The extra-cellular stimulation can be converted to an equivalent model with distributed intra-cellular current sources.

Figure 5.10 Current flow induced by an extracellular current source and polarizationpattern (changes in transmembrane potential) along the cable. A cathodic extracel-lular point-source electrode will depolarize the cable (light) near the electrode andhyperpolarize both sides of the depolarized region (dark).



metric structure), the voltage distribution depends on the cable geometry.Figure 5.11 illustrates the voltage profile along a long dendritic tree (darkline from cell body to the dendritic terminal) of a preganglionic parasympa-thetic neuron with a realistic morphology reconstructed from cat spinal cordby Morgan and Ohara.35 The dendrite diameter changes along the tree andthe transmembrane voltage depends on the cable diameter.

Under extracellular stimulation, the transmembrane voltage responsedepends on the orientation and geometry of the neuron, because the extra-cellular potential outside each compartment is determined by the type ofsource and the distance from it. As seen from the cable equation, the sourcedriving polarization is the second derivative of the extracellular potentialalong the cable and must be non-zero to create changes in membrane poten-tial in the neuron.

As an example the extracellular potential and transmembrane potentialsof three myelinated axons under a point current (I) source were calculatedusing the discrete cable equation (Figure 5.12). The extracellular potential isinversely proportional to the electrode-to-cable distance (R) and is given by:

where σe is the conductivity of the extracellular space with unit of S/cm.The transmembrane potential generated by the point source is triphasic,which is approximately predicted by activating function (see Section 5.2.8).Anodic extracellular stimulation produces maximum hyperpolarization onthe nearest point of the cable and two depolarization peaks next to it, while

Figure 5.11 Simulation result from the real geometry of dendrite. The voltage profilealong the real dendrite is determined by geometric information. This limits theapplication of theoretical cable equation to realistic dendrites or axon where nonlinearion channels are located and activated by this voltage change.

VI

Ree

=4πσ



cathodic extracellular stimulation produces the opposite pattern of polariza-tion. Action potential initiation occurs at the point of maximum depolariza-tion and will thus differ under anodic and cathodic stimulation.

5.2.10 Modeling of synaptic inputs

Excitation of neurons can be initiated by an extracellular electric field asdescribed, but the behavior of the nerve cell is also modulated by synapticinputs from presynaptic terminals that are exposed to the extracellular elec-tric field. Because of the large number of connections, input from presynapticterminals excited by extracellular stimulation can have strong postsynapticeffects (see Section 5.3.2).

Most rapid signaling between nerve cells in the CNS involves iono-tropic transmission that occurs through synaptic connections. When anaction potential arrives at a presynaptic terminal, depolarization resultsin release of neurotransmitters into the synaptic cleft. The released neu-rotransmitters diffuse to the postsynaptic cell membrane where they bindto the receptors. Receptors are activated by specific neurotransmitters andgenerate excitatory postsynaptic potential (EPSP) or inhibitory postsyn-aptic potential (IPSP).

As with other ion channels, activation of postsynaptic receptors changeschannel conductance, which generates a postsynaptic current described by:

Figure 5.12 Transmembrane voltage of axons induced by extracellular current sourceand extracellular voltage along the axon. (A) The magnitude of transmembranevoltages (top traces) is proportional to axon diameter, because the larger diameteraxon has a larger internodal distance, which will increase activating function byraising the second derivative ∆2Ve and conductance term Ga. (B) Extracellular voltagealong the cable is inverse proportional to the electrode to cable distance.

I f g V Esyn syn syn post syn= −( )



where fsyn is fractional transmitter release, is maximal postsynaptic con-ductance, and Esyn is the reversal potential.8,13 A simple form of fsyn is modeledusing an alpha function:42

This function has rapid rise to a peak at tp and slow decay with time constanttp and is a good empirical estimate of the postsynaptic conductance change.As synaptic inputs are modeled as current sources with nonlinear conduc-tances, distribution of synaptic inputs on a modeled postsynaptic neuroncorresponds to adding a forcing term in each compartment. These “indirect”synaptic forcing terms may alter the patterns of neural activation due to the“direct” forcing terms arising from extracellular stimulation and cannot beoverlooked during CNS stimulation.

5.3 Properties of CNS stimulationTo make accurate inferences about anatomical structures or physiologicalmechanisms involved in electrical stimulation, one must know which neuralelements the stimulus pulse activates. When stimulating within the CNS theelectrode is placed within a complex volume conductor where there are threegeneral classes of neurons that can be affected: local cells, axon terminals, andfibers of passage (Figure 5.1). Local cells represent neurons that have their cellbody in close proximity to the electrode. Axon terminals represent neuronsthat project to regions near the electrode and make synaptic connections withlocal cells. Fibers of passage represent neurons where both the cell body andaxon terminals are far from the electrode, but the axonal process of the neurontraces a path that comes in close proximity to the electrode. Each of theseclasses of neurons can be activated by stimulation with extracellular sources.However, activation of each different class can result in different physiologicand/or behavioral outputs. Experimental measurements indicate that localcells, axon terminals, and fibers of passage have similar thresholds for activa-tion when stimulating with extracellular sources (see Section 5.1.1). Therefore,it is often difficult to determine the clear effects of extracellular stimulation.

5.3.1 Direct activation

The seminal review of Ranck45 laid the groundwork for understanding theeffects of electrical stimulation within the CNS; however, because of limita-tions in experimental techniques and the complex response of neurons toextracellular stimulation, our understanding of electrical stimulation of theCNS has advanced at a relatively slow pace. The use of multicompartmentcable models of CNS neurons coupled to extracellular electric fields has givenus the opportunity to address many important issues related to extracellular

gsyn

f ttt

ttsyn

p p

( ) exp( )= −1



stimulation, including the site of action potential initiation and the effects ofchanges in stimulus parameters on activation patterns.18,19,32–34,46,47 The anal-ysis of modeling and experimental findings results in four general conclu-sions regarding the effects of stimulation within the CNS (Figure 5.13):

1. When stimulating local cells with extracellular sources, action poten-tial initiation (API) takes place in a node of Ranvier of the axonrelatively far away from the electrode (Figure 5.13A).

2. When stimulating axon terminals and fibers of passage with extra-cellular sources, API takes place in a node of Ranvier relatively closeto the electrode (Figure 5.13A).

3. Anodic stimuli are more effective in activating local cells than ca-thodic stimuli (Figure 5.13B).

4. Cathodic stimuli are more effective in activating fibers of passageand axon terminals than anodic stimuli (Figure 5.13B).

Figure 5.13 Direct excitation of neurons with extracellular stimulation. (A) Actionpotential initiation (API) and propagation for two different electrode locations (elec-trode-to-neuron distance of 100 µm) and cathodic stimulus pulses with durations of0.1 msec. The left and right columns correspond to the responses from electrodeslocated over the axon or cell body, respectively. Each row shows the transmembranevoltage as a function of time at the segment of the neuron shown to the left. The siteof API is noted by the circled i. (Modified from McIntyre and Grill.32) (B) Input–outputrelations for populations of neurons stimulated by a monopolar electrode. Excitationwas studied using randomly distributed populations of 50 local cells and 50 fibersof passage. Percentages of activated neurons (mean ± one standard deviation fromthree different random distributions) are displayed as a function of stimulus ampli-tude for a monophasic cathodic stimulus (pulse duration [pd] = 0.20 msec) and amonophasic anodic stimulus (pd = 0.20 msec). (Modified from McIntyre and Grill.33)



It should always be noted, however, that activation of any neuron withextracellular electric fields is dependent on four main factors:

1. Electrode geometry and the electrical conductivity of the tissue me-dium — The response of the neuron to stimulation is dependent onthe electric field generated by the electrode, which is dependent onthe size and shape of the electrode. In addition, the inhomogeneousand anisotropic electrical properties of the CNS tissue medium affectthe shape of electric field.21 Therefore, both the type of electrode usedand the region of the CNS where it is inserted will affect the neuralresponse to stimulation.

2. Stimulation parameters — Changes in stimulation parameters canaffect the types of neurons activated by the stimulus and the volumeof tissue over which activation will occur. The four primary stim-ulation parameters are the polarity, duration, and amplitude of thestimulus pulse and the stimulus frequency. In general, alterationsin the stimulus pulse duration and amplitude will affect the volumeof tissue activated by the stimulus, and alterations in the stimuluspolarity and frequency will affect the types of neurons activated bythe stimulus.

3. Geometry of the neuron and its position with respect to the elec-trode — In general, the closer the neuron is to the electrode thelower the stimulation current necessary for activation. However,complex neural geometries such as dendritic trees and branchingaxons result in a large degree of variability in current–distancerelationships (threshold current as a function of electrode-to-neurondistance) of the same types of neurons. Therefore, the orientationof the neural structures with respect to the electrode is of similarimportance to the geometric distance between them, especially forsmall electrode-to-neuron distances.

4. Ion channel distribution on the neuron — Axonal elements of aneuron consist of a relatively high density of action-potential-pro-ducing sodium channels compared to cell bodies and dendrites. Asa result, the axonal elements of a neuron are the most excitable andregulate the neural output that results from application of extracel-lular electric fields. However, while the cell body and dendrites maynot be directly responsible for action potential spiking that resultsfrom the stimulus, they do contain several types of calcium andpotassium channels that can affect neuronal excitability on long timescales when trains of stimuli are used.

5.3.2 Indirect effects

Previous experimental and modeling results have shown that the thresholdfor indirect, or trans-synaptically evoked, excitation or inhibition of localcells stimulated with extracellular sources is similar to (in some cases, depen-



dent on electrode location less than) the threshold for direct excitation oflocal cells (see Section 5.1.1). Indirect excitation or inhibition of local cells isthe result of stimulation-induced release of neurotransmitters that resultsfrom the activation of axon terminals activated by the stimulus. Axon ter-minals are activated at low stimulus amplitudes relative to local cells, espe-cially when cathodic stimuli are used. Therefore, when considering the effectof the stimulus on local cells near the electrode it is probable that a largenumber of axon terminals are activated, resulting in high levels of synapticactivity on the dendritic trees of local cells.

Stimulation-induced trans-synaptic activity can be predominantly exci-tatory, predominantly inhibitory, or any relative mix of excitation and inhi-bition, depending on the types and numbers of synaptic receptors activated.Therefore, the interpretation of the effects the stimulation on the neuronaloutput of local cells is made up of two components: (1) the direct effect ofthe extracellular electric field on the local cell, and (2) the indirect effect ofthe stimulation-induced trans-synaptic excitation and/or inhibition. As aresult, an action potential can be generated either by the stimulus pulse itselfor by indirect synaptic activation. However, activation of axon terminals byextracellular stimuli is nonselective to excitatory or inhibitory neurotrans-mitter release.

In general, the indirect effects of extracellular stimulation of local cellsresult in a biphasic response of a short period of depolarization followed bya longer period of hyperpolarization. This biphasic response is the result ofthe interplay between the time courses of the traditionally fast excitatorysynaptic action and the traditionally slow inhibitory synaptic action. Therole of indirect effects on the output of local cells can be enhanced with high-frequency stimulation (Figure 5.14). If the interstimulus interval is shorterthan the time course of the synaptic conductance, the indirect effects willsummate. Because inhibitory synaptic action traditionally has a longer timecourse than excitatory synaptic action, the effect of this summation is hyper-polarization of the cell body and dendritic arbor of the local cell when high-frequency stimulus trains are used. This hyperpolarization can limit theneuronal output that results from direct effects from the stimulus and canfunctionally block the ability of the neuron to integrate non-stimulation-induced synaptic activity during the interstimulus interval. However,because action potential initiation takes place in the axon of local cells inresponse to the direct effects of the stimulation, local cells will still fire actionpotentials in response to each stimulus pulse given that the stimulus ampli-tude is strong enough (Figure 5.14C).

5.4 Selective stimulationMicrostimulation in the CNS can activate neurons with greater specificity thanis possible with larger electrodes on the surface of the spinal cord or brain.The potential thus arises for electrical activation of intact neuronal circuitry,and, in turn, generation of distributed and controlled physiological outputs



for the study of the neural control of function or for application in neuralprostheses. However, in many regions of the CNS, local cells, axon terminals,and fibers of passage are intermingled in close proximity to the electrode. Ingeneral, only one class of neurons is the target population to achieve thedesired output from the stimulus. Yet, the stimulation used to activate thetarget neurons (e.g., local cells) can also result in activation of the other classesof neurons around the electrode (e.g., axon terminals, fibers of passage). There-fore, techniques that can enable selective activation of target populations ofneurons with little to no activation of non-target populations can improve ourunderstanding of experimental results using electrical stimulation of the CNSand provide important tools for application in neuroprosthetic devices.

Previous modeling and experimental work have shown that local cells havelower thresholds for activation with anodic stimuli, while axonal elements(axon terminals and/or fibers of passage) have lower thresholds with cathodicstimuli (Figure 5.13B);32,33,45 however, chronic application of electrical stimula-tion within the nervous system requires the use of biphasic stimuli because of

Figure 5.14 Neuronal output as a function of stimulus amplitude and frequency. Neu-ronal output (percentage of stimuli in a 500-msec stimulus train that generate propa-gating action potentials in the neuron models) was quantified for direct excitation ofa local cell (A) and a fiber of passage (B) from a train of charge-balanced, cathodic-phase first symmetrical biphasic stimuli (100 µsec per phase). Both neurons had athreshold for activation from a single pulse of 34 µA. (C) Influence of stimulation-induced trans-synaptic inhibition on the neuronal output of the local cell. Each terminalof the presynaptic input was activated by a stimulus, and in turn synaptic conductancesrepresentative of GABAergic inhibition were applied to the dendrites of the local cellfollowing each stimulus in the train. (Modified from McIntyre and Grill.34)



issues related to tissue damage and electrode corrosion.39 In general, whenbiphasic stimuli are used, local cells or axonal elements will be activated duringthe anodic or cathodic phases of the stimulus, respectively, resulting in lowselectivity for activation of a target population (Figure 5.15).33 Alterations in thestimulus frequency and/or stimulus waveform represent techniques that canenable enhanced selectivity of either local cells or axonal elements.

5.4.1 Effect of stimulus frequency

Mammalian neurons exhibit both depolarizing and hyperpolarizing after-potentials that follow an action potential spike. The time course of theseafterpotentials is different in the cell body and axon, and these afterpoten-tials affect the threshold for generation of subsequent impulses.34 Myeli-nated axons exhibit a relatively long-duration (~15 msec), high-amplitude(~5 mV) depolarizing afterpotential (DAP) followed by a long-duration(~80 msec), low-amplitude (~1 mV) afterhyperpolarization (AHP). Neu-ronal cell bodies traditionally exhibit a shorter duration, lower amplitude

Figure 5.15 Selective activation of targeted neuronal populations via alterations inthe stimulus waveform. Using randomly distributed populations of neurons (Figure5.1B), we examined the effect of changes in the stimulus waveform on the relativeactivation of local cells compared to fibers of passage. Plotted is the percentageactivation of local cells as a function of the percentage activation of fibers of passagefor six different stimulus waveforms. The stimulus waveforms (from top to bottomin Figure 5.5; see caption) consisted of a monophasic cathodic pulse (pulse duration[pd] = 0.2 msec), a monophasic anodic pulse (pd = 0.2 msec), a symmetrical anodefirst biphasic pulse (pd = 0.2 msec for each phase), a symmetrical cathode firstbiphasic pulse (pd = 0.2 msec for each phase), an asymmetrical anode first biphasicpulse (pd = 0.2 msec for anodic phase; 0.02 msec for cathodic phase), and an asym-metrical cathode first biphasic pulse (pd = 1 msec for cathodic phase; 0.1 msec foranodic phase). (Modified from McIntyre and Grill.33)



DAP that is followed by a pronounced AHP on the order of ~5 to 10 mVthat reaches its maximum ~10 to 20 msec after the action potential spike.During the DAP of the myelinated axon, the threshold to generate anotheraction potential is decreased, and during the AHP of the neuronal cell bodythe threshold to generate another action potential is increased. The overlapin time course of these afterpotentials for the exploitation of biophysicaldifferences in local cells and fibers of passage to enhance selectivity.

Figure 5.14 shows the responses of a neuron with its cell body near theelectrode compared to the response of a fiber of passage to symmetricalbiphasic stimulus trains.34 Maps of the percent of stimuli that generatedpropagating action potentials during the stimulus train as a function ofstimulus amplitude and frequency were generated. Both the local cell (Figure5.14A) and fiber of passage (Figure 5.14B) can fire in response to the firststimulus in the train when the stimulus amplitude is greater than or equalto 34

µA. However, the near-threshold stimulus amplitudes, the ability ofeither the local cell or the fiber of passage to follow the stimulus train in aone-to-one ratio is affected by the different time courses and amplitudes ofthe afterpotentials in the two neurons.

These results demonstrate that modulation of the frequency of the stim-ulus train can enhance selectivity between activation of cells and fibers ofpassage within the CNS. However, it should be noted that while the selec-tivity of fibers of passage can be increased with high-frequency stimulation,local cells can still respond to the stimulus, albeit at a lower average rate.The limited output of the local cells from high-frequency stimulation couldstill be great enough to generate a functional activation of their efferenttarget, and, conversely, driving fibers of passage over 100 Hz may exceedthe physiological limits for those neurons and result in unexpected orunwanted affects. In addition, the output of local cells at high stimulusfrequencies is dependent not only on its direct excitation characteristics butalso on the role of indirect trans-synaptic influences (Figure 5.14C). There-fore, while modulation of stimulation frequency can enhance selectivity offibers of passage over cells, this technique is not especially effective withoutthe augmentation of alterations in the stimulus waveform.

5.4.2 Effect of stimulus waveform

The stimulating influence of extracellular electric fields is related to thesecond difference of the extracellular potential distribution on the surface ofthe individual neurons, and this stimulating influence will cause regions ofboth depolarization and hyperpolarization in the same cell.32,46,47 In general,when stimulating local cells API occurs in the axon of the neuron, relativelyfar from the electrode, while when stimulating axonal elements (axon ter-minals and/or fibers of passage) API occurs in a region of the fiber relativelyclose to the electrode (Figure 5.13A). Stimulus waveforms can be used thatexploit the nonlinear conductance properties of the neural elements of localcells and axonal elements. Due to geometrical and biophysical factors, API



in local cells takes place in neural elements that are hyperpolarized bycathodic stimuli and depolarized by anodic stimuli.32 As a result, monophasicanodic stimuli are more effective in activating local cells than monophasiccathodic stimuli (Figure 5.13B); however, fibers of passage are stimulatedmore effectively with cathodic stimuli than anodic stimuli (Figure 5.13B).Yet, chronic stimulation requires the use of biphasic stimulus waveforms.39

Therefore, asymmetrical charge-balanced biphasic stimulus waveforms havebeen developed to selectively activate either local cells or axonal elements.33

Stimulus waveforms capable of selectively activating either local cellsor axonal elements use a long-duration, low-amplitude, pre-pulse phasefollowed by a short-duration, high-amplitude, stimulation phase. The long-duration, pre-pulse phase of the stimulus is designed to create a subthresholddepolarizing pre-pulse in the neural element, where excitation will take placein the non-target neurons, and a hyperpolarizing pre-pulse in the neuralelement, where excitation will take place in the target neurons.33 The effectof this subthreshold polarization is to decrease the excitability of the non-target population and increase the excitability of the target population viaalterations in the degree of sodium channel inactivation.20 Therefore, whenthe stimulation phase of the waveform is applied (opposite polarity of thepre-pulse), the target neuronal population will be activated with enhancedselectively compared with monophasic stimuli (Figure 5.15). Further, chargebalancing is achieved as required to reduce the probability of tissue damageand electrode corrosion.

Figure 5.15 shows the effects of changing the stimulus waveform on theactivation of populations of local cells and fibers of passage randomly dis-tributed around the stimulating electrode.33 As seen in Figure 5.13B,monophasic anodic or cathodic stimuli result in selective activation of localcells or fibers of passage, respectively. However, when symmetrical charge-balanced biphasic stimuli are used, selectivity is diminished. Asymmetrical,charge-balanced, biphasic, cathodic-phase first stimulus waveforms result inselective activation of local cells, and asymmetrical, charge-balanced, bipha-sic, anodic-phase first stimulus waveforms result in selective activation offibers of passage. However, even with the appropriate stimulus waveformit should always be noted that selective activation of either local cells oraxonal elements is affected by the stimulation-induced, trans-synaptic influ-ences on the local cells.

5.5 ConclusionElectrical stimulation of the CNS is a powerful tool to study neuronalconnectivity and physiology, as well as a promising technique to restorefunction to persons with neurological disorders. However, the complexityof central volume conductors (brain and spinal cord) and the neuronalelements (cells, axons, dendrites) therein has limited our understanding ofCNS stimulation. To understand the results of studies employing CNSstimulation requires knowledge of which neuronal elements are affected



by stimulation, and optimizing neural prosthetic interventions requirestechniques that enable selective stimulation of targeted neuronal popula-tions. Computational modeling is a powerful tool to address both thequestion of what neuronal elements are activated by CNS stimulation andto design optimal interventions.

AcknowledgmentsPreparation of this chapter was supported by NIH Grant R01-NS-40894.

References1. Altman, K.W. and Plonsey, R., Development of a model for point source

electrical fibre bundle stimulation, Med. Biol. Eng. Comput., 26, 466–475, 1988.2. Anderson, D.J., Najafi, K., Tanghe, S.T., Evans, D.A., Levy, K.L., Hetke, J.F.,

Xue, X., Zappia, J.J., and Wise, K.D., Batch fabricated thin-film electrodes forstimulation of the central auditory system, IEEE Trans. Biomed. Eng., 36,693–698, 1989.

3. Bak, M., Girvin, J.P., Hambrecht, F.T., Kufta, C.V., Loeb, G.E., and Schmidt,E.M., Visual sensations produced by intracortical microstimulation of thehuman occipital cortex, Med. Biol. Eng. Comput., 28(3), 257–259, 1990.

4. Baldissera, F., Lundberg, A., and Udo, M., Stimulation of pre- and postsyn-aptic elements in the red nucleus, Exp. Brain Res., 15, 151–167, 1972.

5. Bannister, N.J. and Larkman, A.U., Dendritic morphology of CA1 pyramidalneurones from the rat hippocampus. II. Spine distributions, J. Comp. Neurol.,360, 161–171, 1995.

6. Bilkey, D.K. and Schwartzkroin, P.A., Variation in electrophysiology and mor-phology of hippocampal CA3 pyramidal cells, Brain Res., 514, 77–83, 1990.

7. Burke, D., Movement programs in the spinal cord [commentary], Behav. BrainSci., 15, 722, 1992.

8. Calabrese, R., Hill, A., and VanHooser, S., Realistic modeling of small neuronalcircuits, in Computational Neuroscience, DeSchutter, E., Ed., CRC Press, BocaRaton, FL, 2001, pp. 259–288.

9. Campbell, P.K., Jones, K.E., Huber, R.J., Horch, K.W., and Norman, R.A., Asilicone-based three-dimensional neural interface: manufacturing processesfor an intracortical electrode array, IEEE Trans. Biomed. Eng., 38, 758–767, 1991.

10. Coburn, B., Electrical stimulation of the spinal cord, two-dimensional finiteelement analysis with particular reference to epidural electrodes, Med. Biol.Eng. Comput., 18, 573–584, 1980.

11. Coburn, B., A theoretical study of epidural electrical stimulation of the spinalcord. II. Effects on long myelinated fibers, IEEE Trans. Biomed. Eng., 32,978–986, 1985.

12. Coburn, B. and Sin, W.K., A theoretical study of epidural electrical stimulationof the spinal cord. I. Finite element analysis of stimulus fields, IEEE Trans.Biomed. Eng., 32, 971–977, 1985.

13. Destexhe, A., Mainen, Z., and Sejnowski, T., An efficient method for comput-ing synaptic conductances based on a kinetic model of receptor binding,Neural Comput., 6, 14–18, 1994.



14. Finley, C.C., Wilson, B.S., and White, M.W. Models of neural responsivenessto electrical stimulation, in Cochlear Implants: Models of the Electrically Stimu-lated Ear, Miller, J.M. and Spelman, F.A., Eds., Springer-Verlag, New York,1990, pp. 55–96.

15. Frijns, J.H.M., de Snoo, S.L., ten Kate, J.H., Spatial selectivity in a rotationallysymmetric model of the electrically stimulated cochlea, Hearing Res., 95, 33–48,1996.

16. Frijns, J.H.M., de Snoo, S.L., and Schoonhoven, R., Potential distributions andneural excitation patterns in a rotationally symmetric model of the electricallystimulated cochlea, Hearing Res., 87, 170–186, 1995.

17. Giszter, S.F., Grill, W.M., Lemay, M.A., Mushahwar, V., and Prochazka, A.,Intraspinal microstimulation, techniques, perspectives and prospects for FES,in Neural Prostheses for Restoration of Sensory and Motor Function, Moxon, K.A.and Chapin, J.K., Eds., CRC Press, Boca Raton, FL, 2001, pp. 101–138.

18. Greenberg, R.J., Velte, T.J., Humayun, M.S., Scarlatis, G.N., and de Juan E., Acomputational model of electrical stimulation of the retinal ganglion cell, IEEETrans. Biomed. Eng., 46, 505–514, 1999.

19. Grill, W.M. and McIntyre, C.C., Extracellular excitation of central neurons,implications for the mechanisms of deep brain stimulation, Thalamus RelatedSyst., 1, 269–277, 2001.

20. Grill, W.M. and Mortimer, J.T., Stimulus waveforms for selective neural stim-ulation, IEEE Eng. Med. Biol., 14, 375–385, 1995.

21. Grill, W.M., Modeling the effects of electric fields on nerve fibers influence oftissue electrical properties, IEEE Trans. Biomed. Eng., 46, 918–928, 1999.

22. Grill, W.M., Electrical activation of spinal neural circuits, application to motor-system neural prostheses, Neuromodulation, 3, 89–98, 2000.

23. Gustafsson, B. and Jankowska, E., Direct and indirect activation of nerve cellsby electrical pulses applied extracellularly, J. Physiol., 258, 33–61, 1976.

24. Hodgkin, A.L. and Katz, B., The effect of sodium ions on the electrical activityof the giant axon of the squid, J. Physiol. (London), 108, 37–77, 1949.

25. Holsheimer, J., Nuttin, B., King, G.W., Wesselink, W.A., Gybels, J.M., and deSutter, P., Clinical evaluation of paresthesia steering with a new system forspinal cord stimulation, Neurosurgery, 42, 541–547, 1998.

26. Jankowska, E., Padel, Y., and Tanaka, R., The mode of activation of pyramidaltract cells by intracortical stimuli, J. Physiol., 249, 617–636, 1975.

27. Kelvin, W.T., On the theory of the electric telegraph, Proc. Roy. Soc. London,7, 382–399, 1855.

28. Lee, D. and Grill, W., Polarization of a spherical cell in a non-uniform electricfield: transient response and comparison with polarization in a uniform field,in press.

29. Lee, D. and Grill, W., Polarization of a spherical cell in a nonuniform electricfield: steady state analysis, Ann. Biomed. Eng., 29(suppl.), S-128, 2001.

30. McCreery, D.B., Shannon, R.V., Moore, J.K., and Chatterjee, M., Accessing thetonotopic organization of the ventral cochlear nucleus by intranuclear micro-stimulation, IEEE Trans. Rehabil. Eng., 6, 391–9, 1998.

31. McDonagh, J.C., Gorman, R.B., Gilliam, E.E., Hornby, T.G., Reinking, R.M.,and Stuart, D.G., Electrophysiological and morphological properties of neu-rons in the ventral horn of the turtle spinal cord, J. Physiol. Paris, 93, 3–16, 1999.

32. McIntyre, C.C. and Grill, W.M., Excitation of central nervous system neuronsby nonuniform electric fields, Biophys. J., 76, 878–888, 1999.



33. McIntyre, C.C. and Grill, W.M., Selective microstimulation of central nervoussystem neurons, Ann. Biomed. Eng., 28, 219–233, 2000.

34. McIntyre, C.C. and Grill, W.M., Extracellular stimulation of central neurons:influence of stimulus waveform and frequency on neuronal output, J. Neuro-physiol., 88, 1592–1604, 2002.

35. Morgan, C.W. and Ohara, P.T., Quantitative analysis of the dendrites of sacralpreganglionic neurons in the cat, J. Comp. Neurol., 437, 56–69, 2001.

36. Nowak, L.G. and Bullier, J., Axons, but not cell bodies, are activated byelectrical stimulation in cortical gray matter. I. Evidence from chronaxie mea-surements, Exp. Brain Res., 118, 477–488, 1998.

37. Nowak, L.G. and Bullier, J., Axons, but not cell bodies, are activated byelectrical stimulation in cortical gray matter. II. Evidence from selective inac-tivation of cell bodies and axon initial segments, Exp. Brain Res., 118, 489–500,1998.

38. Ohara, P.T. and Havton, L.A., Dendritic architecture of rat somatosensorythalamocortical projection neurons, J. Comp. Neurol., 341, 159–171, 1994.

39. Pudenz, R.H., Bullara, L.A., Jacques, S., and Hambrecht, F.T., Electrical stim-ulation of the brain. III. The neural damage model, Surg. Neurol., 4, 389–400,1975.

40. Rall, W. and Rinzel, J., Branch input resistance and steady attenuation forinput to one branch of a dendritic neuron model, Biophys. J., 13, 648–687, 1973.

41. Rall, W., Cable theory for neurons, in Handbook of Physiology, Kandel, E.R.,Ed., American Physiological Society, Bethesda, MD, 1977, pp. 39–97.

42. Rall, W., Distinguishing theoretical synaptic potentials computed for differentsoma-dendritic distributions of synaptic input, J. Neurophysiol., 30, 1138–1168,1967.

43. Rall, W., Time constants and electrotonic length of membrane cylinders andneurons, Biophys. J., 9, 1483–1508, 1969.

44. Raman, I.M. and Bean, B.P., Ionic currents underlying spontaneous actionpotentials in isolated cerebellar Purkinje neurons, J. Neurosci., 19, 1663–1674,1999.

45. Ranck, J.B., Which elements are excited in electrical stimulation of mammaliancentral nervous system: a review, Brain Res., 98, 417–440, 1975.

46. Rattay, F., Analysis of the electrical excitation of CNS neurons, IEEE Trans.Biomed. Eng., 45, 766–772, 1998.

47. Rattay, F., The basic mechanism for the electrical stimulation of the nervoussystem, Neuroscience, 89, 335–346, 1999.

48. Rattay, F., Analysis of models for extracellular fiber stimulation, IEEE Trans.Biomed. Eng., 36, 676–682, 1989.

50. Rinzel, J. and Rall, W., Transient response in a dendritic neuron model forcurrent injected at one branch, Biophys. J., 14, 759–790, 1974.

51. Roberts, W.J. and Smith, D.O., Analysis of threshold currents during micro-stimulation of fibers in the spinal cord, Acta Physiol. Scand., 89, 384–394, 1973.

52. Robertson, J.D., Unit membranes, in Cellular Membranes in Development, Locke,M., Ed., Academic Press, New York, 1964.

53. Rose, P.K., Keirstead, S.A., and Vanner, S.J., A quantitative analysis of thegeometry of cat motoneurons innervating neck and shoulder muscles, J. Comp.Neurol., 239, 89–107, 1985.



54. Roth, A. and Hausser, M., Compartmental models of rat cerebellar Purkinjecells based on simultaneous somatic and dendritic patch-clamp recordings,J. Physiol., 535, 445–472, 2001.

55. Schmidt, E.M., Bak, M.J., Hambrecht, F.T., Kufta, C.V., O’Rourke, D.K., andVallabhanath, P., Feasibility of a visual prosthesis for the blind based onintracortical microstimulation of the visual cortex, Brain, 119(pt. 2), 507–522,1996.

56. Sheasby, B.W. and Fohlmeister, J.F., Impulse encoding across the dendriticmorphologies of retinal ganglion cells, J. Neurophysiol., 81, 1685–1698, 1999.

57. Sin, W.K. and Coburn, B., Electrical stimulation of the spinal cord: a furtheranalysis relating to anatomical factors and tissue properties, Med. Biol. Eng.Comput., 21, 264–269, 1983.

58. Spielmann, J.M., Laouris, Y., Nordstrom, M.A., Robinson, G.A., Reinking,R.M., and Stuart, D.G., Adaptation of cat motoneurons to sustained andintermittent extracellular activation, J. Physiol., 464, 75–120, 1993.

59. Struijk, J.J., Holsheimer, J., van Veen, B.K., and Boom, H.B.K., Epidural spinalcord stimulation: calculation of field potentials with special reference to dorsalcolumn nerve fibers, IEEE Trans. Biomed. Eng., 38, 104–110, 1991.

60. Struijk, J.J., Holsheimer, J., Spincemaille, G.H.G.H., Gielen, F.L., and Hoeke-ma, R., Theoretical performance and clinical evaluation of transverse tripolarspinal cord stimulation, IEEE Trans. Rehab. Eng., 6, 277–285, 1998.

61. Turner, J.P., Anderson, C.M., Williams, S.R., and Crunelli, V., Morphology andmembrane properties of neurones in the cat ventrobasal thalamus in vitro, J.Physiol., 505(pt. 3), 707–726, 1997.

62. Warman, E.N., Grill, W.M., and Durand, D., Modeling the effects of electricfields on nerve fibers: determination of excitation thresholds, IEEE Trans.Biomed. Eng., 39, 1244–1254, 1992.



chapter six

Semiconductor-basedimplantable microsystems1

Wentai Liu, Praveen Singh, Chris DeMarco, Rizwan Bashirullah, Mark S. Humayun, and James D. Weiland

Contents

6.1 Introduction6.2 Building blocks of an implant microsystem

6.2.1 Bidirectional telemetry unit6.2.2 Optical telemetry6.2.3 Inductive/magnetic telemetry

6.2.3.1 Power-transfer efficiency6.2.3.2 Driving efficiency6.2.3.3 Data-modulation bandwidth6.2.3.4 Secondary side power recovery6.2.3.5 Primary side data encoding and secondary

side data recovery6.2.3.6 Radiofrequency telemetry6.2.3.7 Back telemetry

6.2.4 Electro–bio interface6.2.4.1 Sensing interface6.2.4.2 Stimulating interface

6.2.5 Analog/digital circuitry and digital signal processing6.2.6 Hermetic packaging

6.2.6.1 Failures and mechanisms6.2.6.2 Hermetic packaging techniques

6.3 Implant systems6.3.1 Cardiac pacemaker implant6.3.2 Implant for epilepsy

1 This research is supported by grants from NSF, NIH, and Department of Energy.



6.3.3 Cochlear implants6.3.4 Mobility implants6.3.5 Visual prostheses6.3.6 A prosthetic microsystem for an intraocular prosthesis

6.3.6.1 Electrical Perceptions Experiments6.3.6.2 Progress Summary6.3.6.3 Architecture of the Retina-3.55 micro-stimulator

design6.3.6.4 Experimental measurements6.3.6.5 Telemetry circuitry: data link6.3.6.6 Chip implementation

6.4 ConclusionReferences

6.1 IntroductionThe field of integrating a dysfunctional biological subsystem with microelec-tronic systems is an emerging and fast growing one. The main functionalityof the implantable microelectronics are (1) biochemical sensor for sensing ofvital life signs, oxygen, pH, glucose, temperature, and toxins by miniaturizedchemical and bio-sensors; (2) telemetry, to provide a local or remote link fordata/audio/visual signals; and (3) actuation, to provide life saving, to restorelost function, and augment normal function. Implantable and/or wearablemicroelectronics can be used to replace lost function or to monitor physio-logical conditions if the interface with living tissue can be properly achieved.Some success has been obtained, including the cochlear prosthesis, implant-able stimulators for the central nervous system to alleviate pain and reducethe unwanted tremor associated with diseases such as Parkinson’s, the vagusnerve implant for reducing the seizure activity of epilepsy, a functionalneuromuscular implant to restore motor mobility for paraplegia, and thevisual prosthesis for restoring eyesight. Furthermore implantable or wear-able microelectronics could potentially augment the existing human sensesof hearing, sight, touch, smell, and taste beyond the current abilities of thehuman body. This ability would be useful for individuals in hostile environ-ments, such as soldiers and firefighters, or in dedicated situations, such asdoctors performing microsurgery. In particular, we believe very strongly thatjust as novel drug and gene therapies have future roles to play in curinghuman disease so does the field of implantable or wearable microelectronics.

Today, the scale of micron technology is compatible with the cell dimen-sion, while the scale of deep-submicron and nanotechnology is compatiblewith molecular dimensions. Clearly the advances of micro- and nano-fab-rication would greatly benefit the development of the core implantabletechnology. Thus, with the development of ultra-low-power CMOS (com-plementary metal-oxide semiconductor) microelectronics, advances inmicro-fabrication and microelectromechanical systems (MEMS) technology,



these implantable technologies will eventually enable the realization of anintegrated system for stimulation and data collection of electrical and/orbiochemical activity over long time scales in freely moving subjects. Thekeys toward efficient and successful implantable microelectronics are min-iaturization, extremely low power dissipation, biocompatibility, andimplant durability. Implantable devices with smartness and mobility requirean integration of information technology and wireless technology. Thus,implantable microelectronics provide a challenge in material, device fabri-cation, and design technique. In order to capture this apparent wealth ofopportunity, a multidisciplinary team is necessary.

The core implantable technologies include biocompatible materials, inter-face of device and living tissue, miniaturized passive and active devices, her-metic sealing/packaging, energizing mechanism, heat and EM propagationmodeling in living systems, wireless link of distributed sensors and actuators,and signal processing. In this chapter, the issues and perspectives of theimplantable technology have been exemplified by the presentation of theintraocular prosthesis project. The intraocular prosthesis project was conceiveda decade ago and requires a multidisciplinary research effort toward animplantable prosthetic device. The rehabilitative device is designed to replacethe functionality of defective photoreceptors in patients suffering from retinitispigmentosa (RP) and age-related macular degeneration (AMD). In the nextsection, the system building blocks for prostheses are presented.

6.2 Building blocks of an implant microsystemImplant microsystems usually share a common framework: an external sig-nal processing unit for biological sensory information (sound, image, etc), abidirectional telemetry unit, an internal signal processing unit, a stimulusgenerator/driver, and an electrode array for interfacing to tissue or nerves.Figure 6.1 shows a typical implanted microsystem. Various applications

Figure 6.1 Building blocks of implantable microsystems.



require different implementation substrates, such as ceramics, polyimide,thin-film, and silicon; however, the future implanted microsystems wouldbe fully integrated, motivated by the factors of power dissipation, size, andpackaging. Modern silicon technology does offer the advantage of capabilityof integration, thus a silicon implant has the potential of becoming thedominant one in future implant applications. Because the implant is themechanism for interfacing with living tissues, care must be exercised toprevent toxicological damage and infection. Thus, biocompatibility is ofgreat concern and accordingly a hermetic package is a necessity in the assem-bly of an implanted microsystem.

6.2.1 Bidirectional telemetry unit

Developing implantable prostheses are usually partitioned to include someimplanted components and some exterior components, at least until suchtime as the prostheses have well matured. It seems ideal that a prosthesisconsisting of neuro-stimulators or neuro-recorders would have all compo-nents fully implanted such that communication with any exterior devicesis minimized or even eliminated. Accordingly, for biocompatibility, a morestringent power dissipation budget would be necessary to support a fullyimplantable prosthesis. However, a number of advantages can be derivedfrom partitioning a prosthesis into parts internal and external to the body,beyond merely simplifying the implantable unit. Primary benefits of exte-rior electronics, where possible, include the decreased risk of adverse reac-tion to implanted materials, lower internal heat dissipation, and ease ofrefinement and upgrade of signal-processing algorithms and functionalityin the exterior components. This is the case for the NCSU/JHU/USC epi-retinal prosthesis.1

Previously, invasive tethering to implanted electronics using percutane-ous connectors has been used to facilitate the studies of long-term biocom-patibility and the investigation of the necessary signal preprocessing neededto enhance the performance of implants;2 however, in all cases of chronicprostheses, but especially with the ocular prostheses (owing to the eye’sfragility), percutaneous connectors are unsuitable. Therefore, this sectionsummarizes the common types of noninvasive connections to bio-implantsand the current progress of biotelemetry as it relates to the power andcommunication needs of prostheses.

The use of transcutaneous, or wireless, connections is justified due tothe many complications that are typical for percutaneous schemes. Percuta-neous connectors (physically wired links between implanted and exteriorprosthesis components) raise the risk of infections due to a perpetual breachof the body, through which wires must pass. Depending on mechanicalanchoring, such as to bone, percutaneous connectors may restrict movementof a prosthesis in the tissues in which it is implanted and with which itinterfaces electrically. These tissues may be free to move with respect to apercutaneous connector with obvious complications. In the case of ocular



prostheses, rapid eye movement can break any penetrating wires passingthrough the scleral wall, for example, that might otherwise be used to con-nect an intraocular prosthesis to external electronics. Furthermore, dislodg-ing of the implant due to external tethering on the wires could occur. In thecase of most all prostheses, implantable batteries are undesirable because ofthe associated replacement surgery (except where charging can be initiatedfrom outside of the body, as from coils, for example). Even renewable cellshave limited recharge cycles. Furthermore, as a consequence of the biologicalenvironment, implanted batteries must be enclosed in the implant encapsu-lant, which complicates replacement.

A wireless link is a viable approach to support the required data band-width and simultaneously provides adequate power to an implanted pros-thesis. For example, devices functioning as stimulators also require config-uration and stimulus data to function. Furthermore, in the case that theimplant also performs neuro-recording and/or monitors biofunctions ordevice status, or performs self-diagnostics, then there is also the need totransmit data to exterior devices. This functionality is termed back-telemetry.Therefore, in the general case, the success of prostheses for the long termrequires a telemetry link, which can provide adequate power to all implantedelectronics and support a means to communicate information bidirectionally.Three types of telemetry are considered to have the potential of meeting thegoals of simultaneous power and data delivery to implants. These are optical,magnetic (inductive), or electromagnetic (radiofrequency, RF). Each of theseis expounded upon in the following sections.

6.2.2 Optical telemetry

Optical telemetry appears primarily in research groups that are addressingocular prostheses, as the cornea and crystalline lens are usually transparent.This form of telemetry uses a source of high-energy light, such as a laser, toexcite the photoelectronics that form part of the implant. The MIT/Harvardepi-retinal prosthesis group3,4 and the epi-retinal prosthesis group at Fraun-hafer, Germany,5 have been researching optical telemetry.6

The architecture of the first-generation MIT/Harvard prosthesisemployed a photodiode array and stimulator attached to either side of athin-film polyimide electrode array and anchored near the vitreous base. Asupply level of 300 µA at 7 V was derivable with exposure to a 30-mW, 820-nm laser. A second-generation architecture has been proposed withimproved electronics mounted in the crystalline lens position on a polymerannulus. Stimulation data were transmitted via intensity modulation of theexternal laser. No back-telemetry facilities appear to have been provided.

6.2.3 Inductive/magnetic telemetry

Inductive telemetry has been the standard means of wireless connection toimplanted devices for years.7,8 It is based on the mutual magnetic coupling



of two proximal, coaxial, and coplanar coils. The secondary coil in thisarrangement is implanted along with the stimulator/neuro-recorder, whichit services, while the primary coil remains exterior. A low-frequency “carrier’’(i.e., usually 1 to 10 MHz) is driven onto the primary coil, which can thenbe coupled onto the secondary coil for the transfer of power. Furthermore,this power carrier can be modulated to transmit a datastream to the implant.This arrangement forms the basis for inductive coupling, while the precisedetails of how the primary coil is excited and modulated and how DC powerand data are recovered on the secondary (implanted) side are varied andcontinue as topics of research.

Two major disadvantages, however, are noted regarding magnetic cou-pling and are simply characteristics of the coils. These impact the develop-ment of the primary-coil excitation and modulation circuits on the exterior(non-implanted) side:

• Radiation compliance — It is obviously desirable to radiate energyonly toward the secondary coil. In reality, the primary coil radiatesenergy widely in many directions, particularly with no ferrous coreto concentrate the magnetic flux, as in a transformer. This omnidi-rectional radiation pattern imposes unnecessary drain of the sys-tem batteries. It also potentially imparts EMI to other nearby elec-trical devices, and care must be taken to meet the regulatoryguidelines. Inevitably, various strengths of electric and magneticfields are emitted from the primary coil. Thus, deposition of elec-tromagnetic and thermal energy in the proximity of the implant isa serious safety issue in these kinds of telemetry systems. Thedesign of the telemetry link must comply with the IEEE safetystandard with respect to the human exposure to the radiofrequencyelectromagnetic fields.9

• Coupling problem — Due to the implantation, the coils must be disjointat a distance and subsequently cannot be coupled, or linked, by acommon ferrous material but are rather “air-cored.” Therefore, themutual inductance, or coupling coefficient, could be low. High mag-netic field strength in the primary coil is required in order to inducesufficient energy in the secondary coil to power the implant. There-fore, three criteria are used to optimize the inductive link perfor-mance for use in the context of bio-implants: power-transfer efficiency,driving efficiency, and carrier-modulation bandwidth. Each of these issuesis covered separately with attention given to how they have beenaddressed in the research literature.

6.2.3.1 Power-transfer efficiencySeveral studies have been conducted to measure the mutual inductancebetween two air-coupled coils. Coupling strength is dependent on coil load-ing, excitation frequency, number of turns, coaxial alignment, coil separation,coil geometry, and angular alignment. Most of these factors are subject to



variation in prostheses. Typical values for coil coupling are between 0.01 and0.1. Power transfer efficiency can be improved with an increase in the coilcoupling coefficient.10 Power efficiency is achieved most easily when theprimary and secondary coils are series or parallel resonated with a capaci-tance of appropriate value, C = 1/[(2

πf)2L], where L is the self-inductance ofthe coil.

6.2.3.2 Driving efficiencyModulation techniques for data delivery are usually performed using ampli-fier circuits chosen for their high driving efficiency. This efficiency reflectsthe additional power consumption associated with the driving amplifier forenergizing the primary coil. In general, a low coupling coefficient can becompensated by driving the primary coil at high magnetic field strength. Inthis case, it is essential for practical battery life that the driving amplifier ofthe coil be as efficient as possible.

Switch-mode DC–DC converters such as the buck, boost, Cuk, and Sepicachieve a high efficiency by retaining power which is not used by the loadduring each cycle. Rather than dissipating this unused energy as heat, as inlinear or shunt regulators, these switch-mode devices trade the energy backand forth between electric and magnetic fields by resonating the coils. Forinductive links, this also achieves a high magnetic field strength in theprimary coil, while retaining the efficiency.

The circuit topology of choice for driving the resonated primary coil isthe class-E type of amplifier, first characterized by Sokal.11,12 In these originalcontexts, the amplifier is shown to drive a resistance load as illustrated inFigure 6.2a. This amplifier differs from class-A through class-D types ofamplifiers in that the active device operates as a switch rather than as acurrent source.

The class-F amplifier shown in Figure 6.2b also boasts high-efficiencyoperation using the concept of harmonic termination to shunt frequencycomponents, which would cause high-power dissipation in the switch.14

However, it is generally more difficult to design than the class-E amplifierand requires a correctly sized transmission line, as in Figure 6.2b, or else atheoretically infinite number of resonated inductive elements to achieve thehigh-efficiency condition.

Because of the low coupling coefficient, it is expected that minor shiftsin the load impedance on the secondary side will be negligibly reflected ontothe primary and thus will not affect class-E performance significantly. How-ever, it has been pointed out that the close proximity of the primary coil tometallic objects can alter self-inductance and equivalent resistance or defor-mations in coil geometry may heavily disrupt proper class-E operation.15

Therefore, autocompensation of the class-E frequency is proposed to affectchanges in component values (particularly shifts in L2) using a feedbackcontroller to maintain high operating performance. The technique for thisinvolves tapping the magnetic field of L2 with a sensing coil in order tomonitor the zero-crossing of the inductor current.



6.2.3.3 Data-modulation bandwidthMany micro-stimulator designs documented recently in the literature havestimulus instructions encoded into packets for transmission using digitalcommunication protocols.16–20 It is advantageous for advanced signal-pro-cessing hardware to remain outside of the body with only the stimulus unitand associated hardware implanted. Therefore, once a link design has beenestablished that can support the power demands of the implant, capabilitiesfor transmitting the datastream must be added. For a low data rate, this iscommonly accomplished by modulating the power carrier using a conven-tional scheme, such as amplitude, frequency, or phase-shift keying. Forpower carrier modulation, amplitude shift keying (ASK) appears as thedominant modulation scheme for inductive links, as discussed below.

Ziaie et al.21 noted that the voltage developed across inductor L2 usedas the primary inductor is linearly related to the class-E amplifier supplyvoltage, Vdd. This motivates one to perform ASK power-carrier modulationby switching the supply voltage between two levels using the circuit con-figuration of Figure 6.3. Capacitor Cr is used to filter switching noise fromthe amplifier supply. A data bandwidth of 80 kb/sec using this techniqueis reported.

a

b

Figure 6.2 Amplifier topology: (a) class E, (b) class F. (From Lee, T.H., The Design ofCMOS Radio Frequency Integrated Circuits, University Press, Cambridge, U.K., 1998.With permission.)



Troyk and Schwan15 propose a method of amplitude modulating thepower carrier radiated from inductor L2 in the class-E amplifier. By intro-ducing a slight duty-cycle shift into the feedback controller, which pro-duces the gate drive for the switching field effect transistor (FET), the class-E amplifier is moved slightly off of the optimum operating frequency. Asreported, 0.1% changes in oscillation frequency can produce upwards of10% decrease in primary coil current. A disadvantage to power carriermodulation in resonated, high-Q circuits such as the class-E amplifierdescribed by Troyk and Schwan15 is that the forced deviations in operatingfrequency do not track quickly. It is evident from their research that 5 to10 cycles, at best (for 760-kHz operation), are required for the carrier tosettle into a new steady-state amplitude. This imposes constraints onachievable data bandwidth. Another limitation of the ASK modulationdescribed by Troyk and Schwan15 is that in pushing the amplifier awayfrom the class-E frequency, the amplifier is forced away from optimumpower, or driving, efficiency.

This approach is further improved using a technique called suspendedcarrier modulation, which provides a means of performing on/off shift keying(OOSK) with greater achievable bandwidth than with the ASK modulationtechnique described previously. One notable disadvantage of this scheme isthat the implant is not externally powered during carrier suspension andmust draw power from its internal supply capacitance because the primarycoil carries no current during these times. Mueller and Gyurcsik22 havereported another scheme for performing enhanced bandwidth ASK modu-lation of power carriers in high-Q amplifiers. The scheme is called half-cycleamplitude shift keying (HCASK) and is claimed to have a potential for thetransmission of 2 b/cycle of the power carrier.

Figure 6.3 Supply level switching in a class-E amplifier.



6.2.3.4 Secondary side power recoveryOn the receiver side, inside of the body, DC power is usually recovered inconventional ways using a rectifying diode. Secondary coils are either seriesor parallel resonated to maximize power-transfer efficiency from primary tosecondary. Power recovery schemes are designed for single or dual supplylines. Charge-balanced stimulation is used to prevent chronic damage. Out-put drivers of the H-bridge type, as illustrated in Figure 6.4a, only work well

a

b

Figure 6.4 Stimulus circuit architecture types for the (a) bipolar and (b) monopolarelectrode configurations.



when using bipolar electrodes. Although this can be accomplished with asingle supply rail, Vdd relative to Gnd, two electrode contacts (output andindependent return) are required per stimulus output. Monopolar electrodes(shared return) on the other hand, are better suited for use with a positiveVdd rail and negative Vss rail relative to Gnd to actively source and sinkstimulus currents to the load, as shown in Figure 6.4b.

A concern with inductive link designs is that changes in coil couplingand changing load conditions may induce variable and sometime exces-sive voltage at the secondary side. There are three potential solutions tothis problem. The first and less often attempted approach is to monitorthe supply levels and to detune the resonance of the secondary coil tolower the coupling coefficient and power-transfer efficiency. A secondapproach is to instruct the exterior electronics on the primary side viaback-telemetry to lower the power transfer. A third solution is an on-chipsupply regulator.

For example, in Ziaie et al.,21 a parallel resonated secondary coil drivesa half-wave rectifier. This is used with a reversed Zener diode regulator torecover Vdd and Gnd. This may be an undesirable arrangement for sometissue types as the Zener is expected to generate heat as it shunts unwantedpower. Gudnason et al.,24 Von Arx and Najafi,25 and Nardin and Najafi26 usebridge rectifiers with Zener diode regulators. Jones and Normann,17 Tangheand Wise,18 and Kim and Wise19 have used dual-voltage supplies withpush/pull stimulus-circuits for monopolar electrodes. Ziaie21 has improvedupon the use of Zener diodes as regulators, using them to bias a NPN BJTin series with the supply rail.

Conventional P–N diodes used as rectifiers can account for appreciablelosses and heating because of the inherent 0.7-V drop during conduction.Schottky diodes can be used to reduce the power loss as they offer a lowervoltage drop during conduction (typically 0.5 V). The disadvantage withSchottky diodes is that they are not as easily integrated with stimuluscircuits. A further rectification strategy used with implantable devices isthe synchronous rectifier constructed from a metal-oxide semiconductorfield effect transistor (MOSFET).27 With this approach, FETs emulate therectifying function of diodes by switching on and off accordingly with gatecontrol automatically derived from the received carrier. With voltage dropsas low as 0.1 V, the synchronous rectifiers can offer heat reduction by asmuch as 50%.

6.2.3.5 Primary side data encoding and secondary side data recoveryWith a realizable carrier modulation scheme in place that can coexist withthe power transfer to the implant, the issue remains of encoding data fortransmission. Usually implanted micro-stimulator integrated circuits (ICs)do not possess an onboard timing reference; therefore, the asynchronoustransmission of datastreams is not feasible as in established schemes suchas RS232, RS488, USB, etc. The datastreams are typically synchronized to anexternal clock, which must also be supplied to the implant.



Quadrature modulation schemes afford the possibility of modulatingtwo separate carriers with the clock and data signals, respectively. However,this is unattractive because the power carrier may be the only carrier avail-able. Therefore, an alternative approach is to encode the clock and data intoa new digital signal, which is subsequently used to modulate the carrier. Awell-known scheme for performing this is the Manchester encoding algo-rithm, which is trivial to encode (on the primary side) but is more difficultto decode on the secondary side, where the silicon area for circuit imple-mentation is more limited.

Another possible approach is to encode the clock and data into acommon signal through the pulse-width modulation of a digital squaresignal according to the data. This technique has been described by Najafi,28

where the carrier is amplitude modulated with short and long pulses toencode logical “0” and “1” data. The pulse-width modulation approach isalso used in the Retina-3 micro-stimulator of the NCSU/JHU/USC epi-retinal prosthesis.1,16 Positive clocking edges are directly reproduced in theencoded signal. Following each positive edge, subsequent pulse widths aredefined according to the datastream content. Logic “0” data are representedwith a duty cycle of 50% and logic “1” data are represented with dutycycles of 25 and 75%. This encoding ensures a zero DC level in the signal.The NCSU approach to recovering the clock and data waveforms on thesecondary side is a two-stage circuit, consisting first of an ASK demodu-lator followed by a delay-locked loop (DLL) in the second stage (refer toSection 6.3.6.5).

6.2.3.6 Radiofrequency telemetryRadiofrequency telemetry is similar to inductive coupling except that a trav-eling electromagnetic wave is employed with antennas rather that a station-ary magnetic field as in inductive coupling. Work is apparently limited dueto (1) antenna size with respect to carrier frequency, and (2) dielectric lossof human tissues with respect to carrier frequency. Because the human bodyis primarily of water content, the permitivities and conductivities of thetissues will be similar. Table 6.1 provides an indication of how the dielectric

Table 6.1 Dielectric Properties of Selected Eye Tissue: Aqueous/Vitreous Humor

Frequency(Hz)

Conductivity(S/m)

Relative Permittivity

Penetration Depth (m)

1E6 1.501 84 0.41151E7 1.502 70.01 0.13161E8 1.504 69.08 0.046561E9 1.667 68.88 0.027021E10 15.13 57.87 0.0027391E11 77.39 7.001 0.0002305

Note: The properties of the ocular tissues listed here were taken from the Dielectric Propertiesof Human Tissues at http://niremf.iroe.fi.cnr.it/tissprop/, where permittivity and con-ductivity were calculated online for various frequencies of relevance to biotelemetry.



http://niremf.iroe.fi.cnr.it/tissprop/

properties of selected eye tissues vary with frequency. Conductivity is shownto increase with frequency, which means that as the carrier wavelengthdecreases it will penetrate to increasingly shallower depths.

6.2.3.7 Back telemetryBack telemetry refers to the capability of a prosthesis to transmit informationfrom implanted electronics to the exterior of the body. This is commonlyimplemented in one of three ways. In active telemetry, coil-based inductivelinks can again be used. If sufficient transparency allows, as in ocular pros-theses, optical telemetry can be used. As an alternative to active methods,passive telemetry can be used. Our approach has been to develop backtelemetry using the power link itself. This work is summarized below.

6.2.3.7.1 Smart bidirectional telemetry system. Current technologiesfor the telemetry between units external and internal to the human body aregenerally based on low-frequency inductive links, because at low frequenciesthe magnetic field can well penetrate the human body. While suitable forlow data rate applications, this approach has the inherent limitation associ-ated with the Q factor that causes a narrow data bandwidth not capable ofsupporting the stimulation of thousands of electrodes at the image rate of60 frames/sec or more required in the future retinal prosthesis. The imagerate of 60 frames/sec is necessary to obtain a nonflickering image whenstimulating remaining retinal neurons of a blind retina.

To accommodate the data required by the increasing number of elec-trodes while still wirelessly providing system power, the next generationretinal prosthetic telemetry unit currently being developed at NCSU isdesigned to operate as a multifrequency telemetry link under the ISM-bandconstraint. The proposed scheme is a hybrid system that separates the datatransmission from the power delivery by allocating different frequencies forthe power and data carriers. The solution is unique compared to conven-tional RF transceivers because of severe constraints in area, power dissipa-tion, and limited intraocular space for off-chip components such as a localcrystal oscillator. Powering of the system is still through coil coupling, lead-ing therefore to a novel mixed low-/high-frequency system.

The use of a separate data carrier permits power-efficient solutions thatactively use back-telemetry as a method to optimize the overall quality ofthe communication link. This means that, through detectable changes of theinternal unit, the primary unit will obtain valuable information on the instan-taneous power level necessary to the internal unit and possible malfunctionsof stimulating electrodes sensed by impedance variations. In particular, it isimportant to keep the received power fairly constant despite coupling vari-ations. Coil separation variations and misalignments due to the displacementof the extraocular unit and/or eye movement exert a negative effect on theinductive coupling coefficient. If the power transfer of the inductive link ishighly dependent on the relative position of the coils, appreciable losses inpower conversion efficiencies result due to voltage regulation within the



implant. The loss of energy in the implanted device, dissipated as heat, hasa detrimental effect on the retina due to long-term increase of temperature.On the other hand, the power transfer efficiency should not fall below theminimum required to produce a stable voltage supply to ensure properdevice operation.

Figure 6.5 shows the inductive link with back-telemetry and voltage-regulation analog front end. The power transmitter for the inductive linkuses a high-efficiency class-E driver to drive the transmitter coil. Rectificationis achieved via a single-phase rectifier and an off-chip charge storage capac-itor. The voltage regulators generate a dual rail supply of ±7 V for theelectrode stimulators and a 3-V supply for the digital logic. An initial shunt-type regulator provides current regulation and coarse voltage regulation. Aseries-type regulator provides the necessary fine voltage regulation requiredfor subsequent stages. The analog front-end also includes the passive back-telemetry unit for reverse data communication. The back-telemeter uses loadmodulation to transmit binary ASK data on the power carrier. Data detectionon the primary side is accomplished by sensing the current in the inductorand low-pass-filtering the analog waveform.

6.2.3.7.2 Back telemetry operation. Communica t ion f rom theimplanted device back to the extraocular unit is accomplished by using thebackscatter load modulation technique, also known as reflectance modulation.Two techniques are used for reflectance modulation: ohmic modulation andcapacitive modulation. Essentially, both modulation techniques are based onthe concept of changing the loaded quality factor, Q, of the resonant circuitto produce amplitude modulation on the received carrier. The operation ofthe back-telemetry unit is shown in Figure 6.6. In order to produce a detect-able change in amplitude on the primary side, the back-telemeter modulates

Figure 6.5 Block diagram of the inductive link, which consists of a power transmitterwith back-telemetry detection circuitry, as well as secondary unit rectification, telem-etry, and regulation analog front-end circuitry.



the switch to change the load impedance seen by the secondary resonantcircuit, thereby changing its quality factor. This change in Q of the secondaryresonant tank is sensed on the primary side via the change in the reflectanceimpedance, ZR. The change in impedance, in turn, modulates the amplitudeof the RF carrier on the primary side, which can be demodulated for datarecovery by sensing the current in its resonant circuit.

The use of the reflectance impedance technique for back-telemetry bringsabout some effects linked with voltage regulation that requires careful atten-tion. First, in general, the change in Q of the implanted device via ohmic orcapacitive modulation will disturb the voltage in the storage capacitor usedto supply charge to the regulator and the rest of the chip. The voltagevariations must be kept lower than the drop-out voltage of the voltageregulator (series type) to ensure proper functionality of the chip, as shownin Figure 6.6. On the other hand, larger modulation indexes are required tofacilitate the data detection at the external receiver, especially if the couplingcoefficients are relatively low. Larger charge storage capacitors can minimizethe voltage variations at the expense of a lower overall quality factor, reduc-ing the overall efficiency of the inductive link. Second, there is an optimalpower transfer that minimizes the power dissipated as heat in the voltageregulators. If too much power is delivered to the implanted device, thevoltage regulators will “burn” an excessive amount of power. Similarly, ifthe power delivered is too low, the voltage regulators will operate in the“drop-out” region.

To resolve these problems, we are developing a novel closed-loop systembetween the power transmitter and the implanted unit via back-telemetrythat will be used to guarantee optimal power transfer and system function-

Figure 6.6 Operation of the back-telemetry unit.



ing. By sensing the current through the shunt regulator in the implantedunit, it is possible to infer power transfer variations due to the relativedisplacement of the units. By counteracting the power fluctuations via anincrease or decrease of the transmitted power, it is possible to maintain thepower delivered to the implanted unit at an optimal level. Thus, an optimalpower supply regulation will be established despite relative displacementsand misalignments of the coils as well as the circuit malfunction in theintraocular unit. Unlike traditional inductive links, the proposed system willaim to minimize the radiated power while ensuring proper device operation.In addition to regulating power transfer variations due to relative coil dis-placements, the proposed active telemetry link will monitor the on-chippower consumption. For instance, if the implanted device is operating onstandby mode where the power consumption by the internal circuitry is verylow, the overall Q of the system will be high. A high Q implies large inducedvoltages, which result in excessive amount of dissipated power as heat inthe voltage regulators. Ensuring minimum radiated power and heat dissi-pation will be accomplished via the proposed active feedback system. Fur-ther improvements will be achieved by efficient design of voltage regulators,power rectification, analog amplifiers, and on-chip band-gap reference cells.

Preliminary simulation results of the back-telemeter and voltage regu-lation circuitry are shown in Figure 6.7. The reverse telemetry data are usedto modulate the Q of the secondary resonant coil of the implanted device,which gives rise to “bursts” in the 135-kHz power carrier that is detectableon the transmitting coil outside the eye. Unlike typical implementations offorward/reverse telemetry links, the reverse telemetry link can be continu-ously operational because the forward data telemetry is operated at a muchhigher frequency. The energy bursts on the power carrier are isolated fromthe 7/14-V power rails via continuous voltage regulation. Notice that thevoltage regulator provides stable power supply rails of 7/14 V despite the

Figure 6.7 Operation of reverse-telemetry and voltage-regulation units.



internal current consumption variations and sharp on/off loading transitions(i.e., standby mode to fully functional). In particular, the telemetry/regula-tion analog front end is designed to provide high regulation efficiency underworst-case current loading variations and reverse telemetry. Simulationresults show that the non-dropout-mode voltage regulation results in aworst-case peak power supply variation of 219 mV, for a regulation efficiencyof approximately 1.5%.

Because the back-telemeter periodically updates the external unit ofinduced voltage levels to obtain improved power regulation efficiency, theon-chip controller internal to the eye uses a synchronization protocol forresource sharing of the back-telemeter. Following the data synchronizationof the telemetry link, a 12-bit data sequence is continuously sent to updatethe external unit; 2 bits are allocated for the power level diagnostics, and theremaining 10 bits are used to update the electrode impedance levels andchip temperature. The back-telemeter controller is synchronized to the on-chip clock used for the digital logic.

6.2.4 Electro–bio interface

The electro–bio interface is a critical issue and may vary significantly fromimplant to implant. Sensing and stimulation of tissues are two types ofinterfaces, as described in next two sections.

6.2.4.1 Sensing interfaceSensing the biological signal may require invasive or noninvasive methods.An example of noninvasive sensing is the electroencephalogram (EEG), inwhich brain activity is recorded using electrodes placed on the head. Ininvasive interfaces biosensors are implanted through surgery. Currentlyresearch also is directed to make interfaces using cultured neuron probes.Neurons are cultivated over the electronic devices and their activities can berecorded. When implanted, the probe neuron will make synaptic connectionsto the host neuron and will provide very effective signal transfer.

6.2.4.2 Stimulating interfaceElectrodes are usually the interface between stimulus circuit and the biologicalcells. The requirement can vary for implants. For instance, in the retinal pros-thesis project, the number of electrodes vary from 10 to 1000. The diameter ofeach electrode is 50 to 400

µm. Because of limited mechanical support insidethe eye, the array structure has to be light weight and it has to be shaped tofit the curvature of the retina and should exert little or no pressure on the retina.

The material for the electrode should be of high conductivity. It shouldbe able to deliver current at high charge densities without corroding ordissolving in the saline water environment. Charge injection capacity, elec-trochemical stability, and mechanical strength of the electrode are consid-ered. Noble metals such as gold (Au), platinum (Pt), or iridium (Ir) or theircompounds (e.g., IrO2) are generally used because of their chemically inert



properties. Safe maximum charge densities are around 100 to 1000 µC/cm2.Other factors to be considered are shape of electrode, polarizability, andfrequency response. Usually small-area electrodes are preferred for betterresolution and uniform charge density.

6.2.5 Analog/digital circuitry and digital signal processing

Sensors are used to convert mechanical, optical, and chemical signals to anelectrical signal. The input analog signal is converted to digital data forprocessing and/or transmission. Based upon the requirements, analog/dig-ital (A/D) circuitry of varying dynamic range, resolution, and speed is used.Data processing is done at a rate from a few kilobytes per second to a fewmegabytes per second (Table 6.2).28a For example, it is required in the fre-quency analysis of signals in the cochlear implants and the mapping of theinput image to have a set of electrodes in the retinal prosthesis. Complexdata processing may be required for some systems. For example, in motionprostheses for smooth movements of parts, command signals from musclesas well as from numerous other sensors have to be processed and sent tonumerous motion controllers.

A typical system may consist of many analog circuit components suchas filters, amplifiers, D/A circuits, A/D circuits, and current mirrors. Implantsystems are either open loop or closed loop. Feedback may be used overallor locally and may or may not have biological components. The stimulusgenerated from the circuit is usually a voltage or current waveform. Therequired output may vary in shape, amplitude, and frequency. Some of thecircuit design constraints may be a minimum input and output impedance,a specific gain, maximum or minimum power and area, low distortion, andlimited variation on outputs. In many implants, low power circuit is a primerequirement for implanted chips. This is because of limited energy availabil-ity or maximum power dissipation capability in the human body. Failureanalysis is a must to analyze the chances of a failure. An example of failuremay be shorting of the output to the power supply. Safeguards are designedto prevent any damage to the cells in case of a failure.

Table 6.2 Data Channels for Different Implants

Implantable Device RF–Channel Utilization and Data Bit Rate

Heart pacemakers Only during configuration and medical checks; 10 kb/sec acceptable.

Phrenic nerve pacemaker

Continuous but low frequency pulses; tens of kb/sec acceptable.

Current cochlear implants

Continuous; <300 kb/sec by the moment.

Bladder control stimulators

3–4 times/day or continuous; <300 kb/sec.

Hand prosthesis Continuous, as high as possible; 1.25 Mb/sec acceptable.Visual prosthesis Continuous, as high as possible; >2Mb/sec acceptable.



6.2.6 Hermetic packaging

Prosthetic implants need to be protected from the hostile ionic environment ofextracellular fluids for the lifetime of an implant recipient, which at times maybe expected to be 100 years. It should be noted that many of the insulatingmaterials that are commonly used may not be employed in the saline waterenvironment. Flow or presence of electric charge because of electrical excitationcan also facilitate some chemical reactions. Most of the electronics are made ofmaterials that are not compatible with the body. For any electronic implant towork reliably for long time, the implants must be hermetically sealed.

6.2.6.1 Failures and mechanismsAn implant can fail in a number of ways.29 Lead wire insulation failure,encapsulation failure, failure at the interface between materials, substratecorrosion, surface dielectric corrosion, MOS gate contamination, P–N junc-tion contamination, and failure of metal and polysilicon interconnects canbe reasons for failure. The failure mechanism can be chemical attack, disso-lution of coating into the body fluid, condensation of water, mobile ions andother biochemicals along interfacial planes, electrochemical reactions to insu-lators leading to dissolution into body fluids, and electrochemical corrosionof the silicon substrate. Careful accelerated testing of the packing should bedone. Many chemical reactions are possible for packaging degradation. Thedominant chemical reaction and dominant mechanism for a particular reac-tion may change with elevated temperatures. The properties of the packingmaterial may also change with temperature increase.

6.2.6.2 Hermetic packaging techniquesThe hermetic packaging not only should encapsulate the electronics but alsoshould provide connections. Polymers such as silicon and fluoropolymershave been successful in accelerated life testing. Silicon is used to achievehermetic packaging in many implants. Stainless steel and titanium are usedto encapsulate in some of the implants. Polyamide substrates have also beenreported to provide hermetic packaged wires and inductors. Research inprotein polymer coating, electronically conducting polymers (polyaniline),and ionically conducting polymers (polyethylene oxide) is being done. Aparticular technique with electrostatic bonding of a custom-made glass cap-sule and a supporting substrate has been reported.30 This sealed feed-throughtechnology allows the transfer of electrical signals through polysilicon con-ductor lines located at the silicon substrate.

6.3 Implant systemsOver the years, many implants have been developed and some of them havebeen very successful. With the advent of powerful and miniaturized tech-nology, improvement in the old systems and development of more sophis-ticated implants are underway.



6.3.1 Cardiac pacemaker implant

While many people have an abnormal heartbeat, for most of people it isharmless. But, for some people, an irregular heartbeat can be a matter oflife or death. The cardiac pacemaker31 is an electronic device that operatesduring the abnormal period to stimulate the heart. The pacemaker has twoparts: the generator and the leads. At the generator, a battery and infor-mation on the stimulation are stored. The wires from the generator, passingthrough a large vein, are anchored in the heart. Electrical impulses deliv-ered through these wires stimulate the heart. The pacemaker can detectwhen the heart beat falls below a certain rate and start stimulation. Simi-larly it turns stimulation off at a certain set rate. The weight of a modernpacemaker is less than 30 g. Average lifetime of the battery is 7 to 8 yearsand it has to be replaced through surgery. The condition of the implant isregularly monitored by a healthcare professional, and reconfiguration ofdevices is done if required.

6.3.2 Implant for epilepsy

The vagus nerve stimulator (VNS), under the brand name NeuroCyberneticProsthesis, works to inhibit seizures by stimulating the vagus nerve in theneck. The VNS, a small pacemaker-like device, is implanted in the seizurepatient’s chest muscle, just below the skin. It consists of a battery generatorand lead wire. The lead wire runs under the skin to the neck where itattaches to the electrodes, which in turn stimulate the vagus nerve. Thevagus nerve runs signals from the body to the brain. At regular intervals,the VNS sends a small electrical impulse to the brain interrupting seizureactivity. The vagus nerve stimulation increases blood flow in both rightand left thalami, which organize sensory messages to the higher brain.Thus, the seizure activity is subdued through stimulating the thalami andregulating thalamic processing.

6.3.3 Cochlear implants

Deafness can be the result of many causes. Deafness that is the result ofdefects in the outer and middle ear, but not further up in the sound sensingsystem, can be corrected by a hearing aid. Cochlear implants32,33 are used forthose people who do not have sensory hair cells and their hearing loss cannotbe corrected by the use of a hearing aid. The cochlear implant electrodesstimulate the auditory nerve. For people with a defect in auditory nerve,auditory brainstem implants (ABIs) are required; however, ABIs are still inthe developmental stages. Initially, cochlear implants were developed toimitate the lost function of the cochlea. Understanding of sound informationprocessing by the cochlea and brain has contributed to the enhancement inthe implant effectiveness. For example, it has been found that the sound-sensing cells are organized in the cochlea tonotopically. The apex of the



cochlea, which senses low frequency of sound, is very thin and winding,hence not accessible to the implant. But, fortunately, the brain is able to fillin the frequency gap if enough overtones are present.

Several factors have been found to affect the effectiveness of the implant.For example, people with deafness acquired after learning speech and lan-guage have performed better with implants than those acquiring them beforelearning speech and language. The implant is more effective for youngerpeople because their brains are more adaptive to the implants. Brains ofyoung children have been found to make new neuron connections. With thehelp of patients, studies have been conducted to improve the effectivenessof the implants. Some of them are early adapting to music and frequencyshifting. Work is being done to increase the number of stimulus sites,decrease the size of implant, reduce the power consumption, apply MEMStechnology, and provide more signal processing.

6.3.4 Mobility implants

Artificial human body parts have been in use for a long time. They haveprogressed from wood to metal to plastic to composite, lightweight material.Electrical and mechanical systems have been assembled in these artificialparts to duplicate the motion of the human body.34 Lately, attempts havebeen made to interface these prosthetic devices with better human control.While sophisticated robotic parts have been manufactured, communicationbetween the parts and humans has been the weakest link.

This communication interface requires the expression of control by theperson in some part of the body, sensing that expression, processing theexpression, and subsequently controlling the artificial parts. While some-times the expression can be easily detected, surgery may be used to implantthe sensors in an alternate region. The complexity of processing the infor-mation and controlling the motion of these robotic parts poses a big chal-lenge. Smooth movement, as in the human body, requires sensing and con-trolling at a large number of points. While full success has not been achieved,steady progress is being made.

Controlling robotic parts directly through the brain is showing greatpromise. The brain commands are sensed noninvasively with an electroen-cephalogram or an implanted electrode. Being noninvasive, the EEG is attrac-tive but for the current technology it provides a very slow rate of information(25 bits per minute) and hence is unsuitable for complex movement control.Implanted chips can gather more information but are too bulky to providea sufficient number of electrodes.

6.3.5 Visual prostheses

Blindness robs millions of individuals of their keen sense of vision. Severalgroups around the world are evaluating the feasibility of creating visualperception by electrical stimulation of the remaining retinal neurons in



patients blinded by photoreceptor loss.35–39 Several neuro-stimulator deviceshave been designed and fabricated for electrical stimulation of tissues. Thecortical prosthesis group at the University of Utah has developed a stimulatorthat accompanies a penetrating electrode array for eliciting visual sensationsthrough cortical stimulation. The neuro-prosthesis group at the University ofMichigan has also produced a number of stimulation devices.18 Additionalwork is in progress among various other researcher groups toward the devel-opment of retinal prostheses.5,39–42 Although it is good to compare theseapproaches, for the sake of brevity we are discussing here a typical approachrepresented by our project. The Retinal-Prosthesis Group at North CarolinaState University, Johns Hopkins University, and University of Southern Cal-ifornia has developed several generations of stimulation ICs. These ICs aredesigned to deliver the currents for retinal stimulation, as were determinedby clinical studies conducted on the visually impaired with RP and AMD.39,43

6.3.6 A prosthetic microsystem for an intraocular prosthesis

The human eye with its nearly 100 million photoreceptors is a complex systemthat allows us to enjoy high-resolution imagery full of colors. The systemfunctions when light is focused by the cornea and crystalline lens within theeye on to the retina. The retina is highly structured, composed of multiplecell layers, each with a specific function. The photoreceptor cell layer captureslight energy and converts it into electrochemical signals. These signals thenexcite bipolar cells located in the second layer of neurons in the retina. Thebipolar cells subsequently transmit the signal to retinal ganglion cells whoseaxons collectively form the optic nerve that connects the retina to the visualcenters of the brain. Other cells such as the horizontal and amacrine cellsform connections between neurons within a layer, adding to the computa-tional power of the retinal network of neurons. This biological neural networkis a sophisticated image processor, capable of compressing information from100 million photoreceptors into 1 million retinal ganglion cells.

While various causes for blindness exist, AMD is the leading cause ofblindness in individuals 60 years or older, with 200,000 pairs of eyes leftlegally blind each year. RP has an incidence of 1 in 4000 and in the UnitedStates alone afflicts 100,000 people. Currently, no treatment exists for eitherRP or AMD. We have shown that in these diseases, despite near total lossof the photoreceptors, the remaining retinal neurons remain intact andpatients could regain vision. It is proposed that artificial vision be providedby the prosthetic system with multiple-unit artificial retina chipset (MARC)implanted within the eye. The system would have an external camera(mounted in a glasses frame) to acquire an image and convert it into anelectrical signal. This signal would be wirelessly transmitted to an implantedchip, which would electrically stimulate the cells of the retina that have notdegenerated. The pattern of electrical stimulation would be controlled toproduce the perception of an image; hence, the sense of vision could bepartially restored with such an implantable electronic device.



6.3.6.1 Electrical Perceptions ExperimentsIn 15 human volunteers who had RP and AMD, intraocular prostheses weretested. It was shown that the remaining neurons in a non-seeing eye can beactivated by an electrical signal, resulting in the perception of light. Thehuman tests were performed in an operating room, where the volunteerswere under local anesthesia. At first tests were done with a single electrode,and then patterns of electrical signals generated by an external computerwere applied to the retina via the electrode arrays. Using a 25-electrode array,at first rows and columns of electrodes and then patterns of electrodesoutlining a large letter or simple geometric shapes such as a rectangle werestimulated. In all experiments it took several minutes before the patient couldconfidently identify the artificially created spot of light. Once they recog-nized the first dot of light they made quick progress. As expected, the 25-electrode array provided poor resolution; however, it was sufficient for allthe patients tested to date to see forms such as a large letter and simplegeometric shape. It was also established that flicker-free vision could beachieved with sufficiently high applied pulse rate. Our results from tests inblind human volunteers demonstrate that form vision is possible using con-trolled pattern electrical stimulation of the retina.

6.3.6.2 Progress SummaryConcomitant with the human tests, we have been engineering a MARC, acompletely implantable system capable of creating hundreds of individualspots of stimulus. Specifically, we had to develop a chip that would besmall enough to be placed within the eye and yet could stimulate the retinathrough a large number of electrodes. The currently developed chip Retina3.55 is 4.6 mm × 4.7 mm and can stimulate 60 electrodes with a safe andeffective charge, and the design of a 1000-channel stimulator is progressing.A prototype image-acquisition system using a miniature CMOS cameraand an image-processing system using a very-large-scale integration (VLSI)image-processing board have been built. Wireless communication electron-ics using a radiofrequency link have been developed that can supply therequired data transmission rate and power coupling. Stimulating electrodematerials are being tested to assess their ability to withstand corrosionwithin the eye.

Of equivalent importance to the engineering development of the deviceis the biological testing that will show that a retinal implant is well toleratedin the eye. Although intraocular retinal surgical procedures are commonlyperformed there is no precedence to attaching a device to the delicate, paper-thin retina. The device could tear the retina, cause intraocular bleeding, ordislodge because of eye movements. A metal alloy tack is used to secure theimplant. In animal experiments, the implants have remained affixed to theretina through 11 months and caused no structural of functional damage.Experiments in progress will examine the effects of electrical stimulation onthe retina.35



6.3.6.3 Architecture of the Retina-3.55 micro-stimulator designFunctional specifications for chip operation require excitation currents up to600 µA amplitude delivered to retinal tissue with a characteristic impedanceof approximately 10 kΩ.44 Variable stimulation rates of 50 to 60 Hz and higherare desired to achieve flicker-free vision as well as definable pulse widthsand inter-phase delays on the order of 1 to 5 msec. Stimulation pulses areproduced in the standard biphasic fashion with anodic-leading or cathodic-leading charge-balanced pulses for compatibility within the biological envi-ronment. To achieve greater flexibility in programming the stimulus sched-ule, the 60 current driver circuits are independently programmable andconnect directly with 60 output channels (or pads). This provides a singlestimulation frame, and thus an intended visual experience in the form of an8 × 8 pixelated image (minus the four corners).

6.3.6.3.1 Communication protocol and subsystems. The instructions thatspecify the operation of the stimulator are defined digitally and loaded intothe IC serially through a data input and an accompanying clock input. Aconfiguration frame packet format is defined for specifying the full-scale out-put current magnitude and stimulus timing, including pulse widths andinterphase delay. Another packet format, designated as the data frame, isdefined for specifying desired stimulation current amplitudes per phase, orfor requesting charge cancellation, for all driver circuits in real time, therebyconstituting a single image frame. Both of these frames are 1024 bits in length,which is convenient for delimiting frame boundaries as packets are loaded.A unique 16-bit synchronization word identifies the beginning of a config-uration frame. If the cyclic redundancy check (CRC) and checksum signa-tures are verified, the configuration data are latched into internal registers.Otherwise, the data are ignored and subsequently overwritten as new pack-ets are shifted into the IC. As with the configuration packet, a data packetis initiated with a unique 16-bit synchronization word, checked, and latched.

6.3.6.3.2 Error-detection subsystem. Data errors could have unin-tended consequences, such as generating biphasic current signals which arenot properly charge balanced. Due to the risks associated with processingerroneous data, CRC and checksum error detection mechanisms are used.A 32-bit cyclic redundancy check (CRC) computation engine is used. Toincrease the robustness, a 16- bit checksum error detection is also used. Thechip will not process any runtime data until a configuration packet is firstreceived without error and latched, as the timing profile will not be defineduntil then. Retina-3.55 does not have back-telemetry at this time to supportthe communication of status or other data back to the extraocular hostcontroller; therefore, the external system cannot know when error-free con-figuration has been achieved. The interim solution to this is retransmissionof configuration-packets at a regular interval. If a valid configuration hasbeen latched, then subsequent error-free runtime data packets will be latchedfor stimulus generation.



6.3.6.3.3 Pulse profile generator design. The pulse profile generatorsproduce eight global, independently programmable (via a configuration-packet) pulsed timing references. These exist in the form of 8-bit (start time,stop time) pairs. Each parameter represents a count of clock cycles, relativeto the start of each of data frame. The generation of the pulse profiles is doneby a counter and comparator circuit, which compares the counter outputwith the start time and stop time.

6.3.6.3.4 Stimulus current driver design. Each of the 60 current driversis capable of producing a biphasic stimulus current using information pro-vided in a 16-bit data subpacket as shown in Figure 6.8. A conceptual modelof the biphasic current driver is as shown in Figure 6.4b. The anodic currentis sourced to the load by an ideal current source referenced to a positivesupply of Vdd. Similarly, the cathodic current is sunk from the load by anideal current source referenced to a negative supply of Vss. The electrodearray44 might be attached to the retina using retinal tacks. Due to implemen-tation in the N-well process, to minimize body effects all digital logic circuitsand low-voltage analog circuits operate between Vss (logic “0”) and Gnd(logic “1”). Vdd is used only in the output stage of the driver circuit. Thedriver employs two NMOS binary-weighted current-mode digital-to-analogconverters (DACs) to produce the anodic and cathodic currents. The DACdesign is as shown in Figure 6.9.

To date, the standard DAC arrangement for producing biphasic currentsconsists of a stacked PMOS-DAC/NMOS-DAC configuration driving amonopolar output.18,19 This structure may not support the high voltagerequired in our design and may require alternative transistor structures thatcan support high drain-source voltages without modification of the process,such as high-voltage transistors.45,46 Therefore, the current-driver structureis modified so that the DAC currents are mirrored and scaled into an outputcircuit of high-voltage compliance. The circuit for this is shown Figure 6.10.Each current driver contains two 8-to-1 multiplexers for programmable selec-

Figure 6.8 Retina-3.55 driver subpacket format.



Figure 6.9 Current-mode DAC circuit detail.

Figure 6.10 Circuit-level detail of the current driver.



tion of any of the eight pulse timing profiles. One component of the drivercircuit is a charge cancellation mechanism, which is intended to limit anyunintentional accumulation of charge on the electrode.

6.3.6.3.5 DAC biasing circuit. The wide-swing cascode current mir-rors in the NMOS DACs and the biphasic amplifier stage receive their biasvoltages from a central biasing circuit. The schematic of the bias circuit isprovided in Figure 6.11. The biasing circuit can be considered in three sec-tions, which are labeled in Figure 6.11 as the bootstrap reference, the DACbiasing section, and the biphasic amplifier section. The bootstrap reference isa circuit that settles into the intersecting operating point of a linear resistanceand a nonlinear transconductance.47 The devices are sized to achieve a 20µA master reference current, which becomes the basis for other currentsproduced on-chip. The DAC biasing section taps the mirrored current of thebootstrap reference to establish two biases for the two NMOS DACs in thedrivers. Using two bits from a prior latched configuration packet (Iref1, Iref0),the DAC biasing section can be tuned to mirror the bootstrap reference withgains of 1×, 2×, or 3×. This ultimately permits global full-scale current outputin the driver’s biphasic amplifiers to be programmed at 200, 400, or 600µA.The driver DACs employ wide-swing cascade mirrors using the associatedbiases of dcas (DAC cascode bias) and ibias (DAC current-source bias).

6.3.6.4 Experimental measurementsThe Retina-3.5 stimulator IC, which consists of the design described hereinwithout the CRC and checksum error detection logic, was implemented in theAMI 1.2-µm CMOS and occupies an area of 4.7 mm × 4.6 mm. As Retina-3.55has not been fabricated at this time, experimental measurements are takenfrom the Retina-3.5 IC design, containing the same stimulus driver core. The

Figure 6.11 Central-bias circuit for DACs and biphasic amplifier.



driver circuit is characterized in terms of linearity and pulse amplitude match-ing, current sensitivity to power supply variations, and effectiveness of thecharge cancellation mechanism and power consumption requirements.

6.3.6.4.1 Output flexibility and example waveforms. A typical biphasicstimulus current generated from Retina-3.5 is shown in Figure 6.12. For thisfigure, the chip is operated from +5 V/–5 V rails and is driving a 10-kΩ loadreferenced to ground. This is a transient plot of the voltage across the loadresistance. The stimulator IC is processing 100 frames per second in thisexample. Figure 6.13a,b illustrates how the anodic/cathodic pulse ampli-tudes can be independently specified. It should be pointed out that in orderto generate a targeted current amplitude of 600 µA, the supply voltages mustbe increased beyond +5 V/–5 V (assuming a 10-kΩ characteristic load imped-ance). Although pulses are achievable at +7 V/–7 V, the matching betweenanodic and cathodic pulses degrades due to limitations in the output stagedesign for such high supply voltages in the AMI 1.2 µm CMOS. Thesesupplies are beyond those rated for the process; however, the gate oxidesand drain/source junctions have been found to be tolerant of +7 V/–7 Vrails without permanent damage.

6.3.6.5 Telemetry circuitry: data link

6.3.6.5.1 ASK demodulator. The prototype implantable devicereceives both power and data via an inductive link. Amplitude modulationinstead of frequency modulation is chosen in order to reduce the circuitcomplexity and power consumption and is able to sustain the data rate ofthe requisite functions. The ASK modulation scheme governs the amplitude

Figure 6.12 Charge-balanced biphasic pulse produced from Retina-3.5.



of the carrier signal according to the desired digital data. Because the recov-ered power is also derived from this amplitude, the average transferredpower could depend on the transmitted digital data pattern. To avoid thisdata dependency, we first encode the data to be transmitted using the alter-nate mark inversion pulse-width modulation (PWM) scheme, which subse-quently modulates the power carrier as shown in Figure 6.14. The system isdesigned for PWM data ranging from 25 to 250 kb/sec with a carrier fre-quency ranging from 1 to 10 MHz.

a

b

Figure 6.13 Independent amplitude variability of the (a) anodic phase and (b) ca-thodic phase.



On the receiver side, the power carrier is rectified and filtered toobtain the base-band carrier envelope containing the PWM data. Theenvelope is further filtered in order to provide a low-ripple DC voltagefor the chip power supply. The unfiltered carrier envelope is passed tothe ASK demodulator circuit of Figure 6.15 to extract the digital (rail-to-rail) PWM waveform.

The demodulator is a comparator with a predefined amount of hyster-esis in which one input is derived from the envelope of the modulatedcarrier (A in Figure 6.15). The other input is derived from the average ofthat signal (B in Figure 6.15). Transistors M0 and M1 provide level shift ofthe input signal to the common mode range of the differential amplifier.

Figure 6.14 PWM and ASK waveforms.

Figure 6.15 ASK demodulator circuit.



Transistor M2 with the 10 pF capacitor C0 serves as a low-pass filter andprovides the average of an input signal. Both signals are applied to thedifferential amplifier with the cross-connected active load of M3 to M6. Thehysteresis is achieved via positive feedback. The demodulator is designedto process the carrier envelope with a ripple between 6.5 to 7.5 V. It has ahysteresis of 500 mV, so as to ensure that the likelihood of an extra tran-sition in the output waveform due to noise is minimized. The outputwaveform of the demodulator is the digital PWM with voltage swingbetween 0 and 7 V.

6.3.6.5.2 Clock and data-recovery circuit. The rising edges of the PWMsignal from the demodulator are fixed in time periodically and serve as areference in the derivation of an explicit clock signal. On the other hand, thedata are encoded by the position of the falling transition at each pulse. Inan alternate mark inversion encoding scheme, a zero is encoded as a 50%duty-cycle pulse, and ones are alternately encoded by 40 or 60% duty-cyclepulses. This eliminates the need for a local clock oscillator such that the clockand data recovery circuit is simplified. It also provides an average coupledpower which is essentially independent of the data.

A diagram of the clock and data-recovery unit is shown in Figure 6.16.It consists of a delay-locked loop (DLL) and decoder logic. The DLL consistsof a phase-frequency detector (PFD), a charge pump, a loop filter, and avoltage-controlled delay line. The 36-stage delay line is locked to one periodof the PWM waveform. The waveform is decoded by an XNOR gate, theinputs of which are the tapped-out signals at the 15th and the 21st stages.The positions for the tapped signals correspond to the duty-cycle percentagesused in the PWM waveform.

A challenge for the DLL design is to make the lockable frequency aslow as 25 kHz without consuming a large chip area. A three-state PFDis used with the additional delay at the reset path to reduce the dead-zone effect. The voltage-controlled delay line is based on current-starvedinverters made of low W/L transistors. The charge pump current is 3.5µA provided by matched wide-swing current sources. A unity gain ampli-fier is included to prevent the ripple distortion of the control voltage dueto charge sharing. The 200 pF loop filter capacitor is integrated on the

Figure 6.16 Clock and data recovery.



chip. The initial condition of the DLL must be set up correctly to guar-antee a correct locking condition. The very first input to the PFD afterreset must be the rising edge of the PWM waveform fed back from thedelay line. This can be enforced by using a simple pulse swallow circuitwhich would only ignore the very first pulse of the incoming PWMwaveform prior to entering the PFD. The loop filter capacitor is initializedto full charge when power is applied. Consequently, the delay line startswith the smallest delay. The loop then initiates the discharge of thecapacitor, resulting in a decrease of the control voltage and an increaseof the delay. This process continues until the delay is exactly equal toone period of the incoming PWM waveform. In this way, locking to asubharmonic of the PWM waveform is prevented. The DLL is alwaysstable because it is a first-order system.

6.3.6.5.3 Communication circuit. The chip is designed only to receiveexternally initiated communication; no back-telemetry is implemented.Although intended for a data rate of 25 to 250 kb/sec, our measurementsshow that the ASK and PWM demodulator circuits can operate in excess of1 Mb/sec. However, at this high data rate, the modulation index needs tobe increased up to 30%. Furthermore, if the ratio of carrier frequency to thedata rate is more than 40, they can be independent of one another. Figure 6.17shows the measured communication waveforms; Ch1 is the carrier envelope(top waveform), Ch2 is the PWM output from the demodulator, and Ch3 isthe recovered non-return to zero (NRZ) data. Note that there are two clockperiods of latency between the NRZ output and the PWM input, owing tothe flipflops in Figure 6.16.

Figure 6.17 Measured communication waveforms.



6.3.6.6 Chip implementationA die photograph of the Retina-3.55 IC is shown in Figure 6.18. The chipwas implemented in the AMI 1.2-µm, two-metal, two-poly CMOS processthrough the MOSIS facility, with die physical dimensions of 4.6 mm × 4.7mm. A peripheral ring of 104 100-µm × 100-µm pads encloses the stimulatorcore. Power and control signal pads are allocated along the bottom edge ofdie to facilitate simple connection, with a separate adjacent IC (not discussedhere) providing power carrier rectification and filtering and supply regula-tion. (Continued development of the stimulator design in context of theinductive powering scheme and wireless telemetry would likely see thecarrier rectifier and filtering and supply regulation co-integrated with thestimulator circuits.) The 60 stimulus-current output pads are distributedalong the remaining three sides of the die.

6.4 ConclusionThe field of semiconductor-based implants is becoming more and moreexciting. Today, advanced semiconductor technologies offer devices at themicron level virtually at the same or smaller size than the biological cell size,the complexity of system on a chip is in the millions of devices, and thespeed of devices can be in the range of several hundred megahertz. More-over, further reduction in device size and advances in the field of microelec-tromechanical systems (MEMS) will help to further miniaturize the implants.Improvements in low power circuit design and higher digital signal process-ing capability are desired. It is also to be noted that projects such as retinalprostheses require expertise in many fields, including chemistry, electronics,materials, and biomedicine. Having the capability to produce sophisticated,low-risk medical procedures for implants and people’s willingness to adopt

Figure 6.18 Fabricated chip (Retina 3.55).



high-tech implants are the driving forces behind new implant development.While cochlear implants, pacemakers, artificial limbs, and portable dialysishave become stable technologies, work on projects such as retinal prostheses,brain implants, and neural recording is in progress.

References1. Liu, W., McGucken, E., Vichienchom, K., Clements, M., de Juan, E., Jr., and

Humayun, M.S., Dual unit visual intraocular prosthesis, Proc. of the 19th Int.Conf. of the IEEE Engineering in Medicine and Biology Society, 1997, pp.2303–2306.

2. Rucker, L. and Lossinsky, A., Percutaneous connectors, 30th Neural ProsthesisWorkshop, NINDS, NINCD, NIH, October 12–14, 1999.

3. Wyatt, J.L. and Rizzo, J.F., Ocular implants for the blind, IEEE Spectrum, 33,47–53, 1996.

4. Rizzo, J.F. and Wyatt., J.L., Prospects for a visual prosthesis, Neuroscientist, 3,251–262, 1997.

5. Eckmiller, R., Learning retina implants with epiretinal contact, OphthalmicRes., 29, 281–289, 1997.

6. Gross, M., Buss, R., Kohler, K., Schaub, J., and Jager, D., Optical signal andenergy transmission for a retina implant, Proc. First Joint BMES/EMBS Conf.,1, 476, 1999.

7. Zierhofer, C.M. and Hochmair, E.S., High-efficiency coupling-insensitive tran-scutaneous power and data transmission via an inductive link, IEEE Trans.Biomed. Eng., 37(7), 716–722, 1990.

8. Zierhofer., C.M., A class-E tuned power oscillator for inductive transmissionof digital data and power, Proc. 6th Mediterranean Electrotechnical Conf., 1,789–792, 1991.

9. IEEE Standard for Safety Levels with Respect to Human Exposure to Radio Fre-quency Electromagnetic Fields, 3KHz–3GHz, IEEE Standard C 95.1., 1999 ed.

10. Zierhofer, C.M. and Hochmair, E.S., Coil design for improved power transferefficiency in inductive links, Int.Conf. Eng. Med. Biol., 4, 1538–1539,1997.

11. Sokal, N.O. and Sokal, A.D., Class-E: a new class of high-efficiency tunedsingle-ended switching power amplifier, IEEE J. Solid-State Circuits, SC-10(3),1975.

12. Sokal, N.O., Class-E: high-efficiency switching-mode tuned power amplifierwith only one inductor and one capacitor in load network — approximateanalysis, IEEE J. Solid-State Circuits, SC-16(4), 1981.

13. Lee, T.H., The Design of CMOS Radio Frequency Integrated Circuits, Universityof Cambridge Press, Cambridge, U.K., 1998.

14. Raab., F.H., Class-F power amplifiers with maximally flat waveforms, IEEETrans. Microwave Theory Techniques, 45(11), 2007–2012, 1997.

15. Troyk, P.R. and Schwan, M.A.K., Closed-loop class E transcutaneous powerand data link for MicroImplants, IEEE Trans. Biomed. Eng., 39(6), 589–599,1992.

16. Clements, M., Vichienchom, K., Liu, W., Hughes, C., McGucken, E., DeMarco,C., Mueller, J., Humayun, M., de Juan, E., Jr., Weiland, J., and Greenberg, R.,An implantable neuro-stimulator device for a retinal prosthesis, Int. Solid-State Circuits Conf., February 1999, pp. 216–217.



17. Jones, K.E. and Normann, R.A., An advanced demultiplexing system forphysiological stimulation, IEEE Trans. Biomed. Eng., 44(12), 1210–1220, 1997.

18. Tanghe, S.J. and Wise, K.D., A 16-channel CMOS neural stimulating array,IEEE J. Solid-State Circuits, 27(12), 1819–1825, 1992.

19. Kim, C. and Wise, K.D., A 64-site multishank CMOS low-profile neural stim-ulating probe, IEEE J. Solid-State Circuits, 31(9), 1230–1238, 1996.

20. Liu, W., Vichienchom, K., Clements, M., DeMarco, S.C., Hughes, C., McGuck-en, E., Humayun, M.S., de Juan, E., Jr., Weiland, J.D., and Greenberg, R., Aneuro stimulus chip with telemetry unit for retinal prosthetic device, IEEE J.Solid-State Circuits, 35(10), 1487–1497, 2000.

21. Ziaie, B., Nardin, M.D., Coghlan, A.R., and Najafi, K., A single-channel im-plantable microstimulator for functional neuromuscular stimulation, IEEETrans. Biomed. Eng. 44(10), 909–920, 1997.

22. Mueller, J.S. and Gyurcsik, R.S., Two novel techniques for enhancing power-ing and control of multiple inductively-powered biomedical implants, Proc.Int. Symp. Circuits Systems, 1, 289–292, 1997.

23. Jun, Z.X. and Hong, Z., A full-custom-designed receiver-stimulator chip foramultiple-channel hearing prosthesis, Proc. Int. Symp. on VLSI Technology,Systems, and Applications, 1993, pp. 288–291.

24. Gudnason, G., Bruun, E., and Haugland., M., An implantable mixed ana-log/digital neural stimulator circuit, Proc. Int. Symp. Circuits Systems, 5,375–378. 1999.

25. Von Arx, J.A. and Najafi, K., A wireless single-chip telemetry-powered neuralstimulation system, IEEE Int. Solid-State Circuits Conf., 1999, pp. 214–215.

26. Nardin, M. and Najafi., N., A multichannel neuromuscular microstimulatorwith bi-directional telemetry, 8th Int. Conf. Solid-State Sensors and Actuators,Eurosensors IX, Transducers Õ95, 1, 59–62, 1995.

27. Matsuki, H., Yamakata, Y., Chubachi, N., Nitta, S.I., and Hashimoto, H.,Transcutaneous DC–DC converter for totally implantable artificial heart usingsynchronous rectifier, IEEE Trans. Magnetics, 32(5), 5118–5120, 1996.

28. Von Arx, J.A. and Najafi, K., A wireless single-chip telemetry-powered neuralstimulation system, IEEE Int. Solid-State Circuits Conf., 1999, pp. 214–215.

28a. Marin, D., Troosters, M., Martinez, I., Valderrama, E., and Aguilo, J., Newdevelopment for high performance implantable stimulators: First 3 Mbps upto 4.4 Mbps demodulator chip through a wireless transcutaneous link, Mi-croNeuro ‘99, pp. 120–126.

29. Edell, D.J., NIH supported work. 1996 #Final report.30. Cheng, L. and Najafi, K. A hermetic glass-silicon package formed using lo-

calized aluminum/silicon–glass bonding, J. Micromech. Syst., 2001.31. Machines in our hearts: the cardiac pacemaker, the implantable defibrillator,

and American Healthcare.32. Cochlear Implants in Adult and Children, NIH consensus statement, May 15–17,

1995, pp. 1–30, http://text.nlm.nih.gov/nih/cdc/www/100txt.html.33. Spelman, F.A., The past, present and future of cochlear prostheses, IEEE Eng.

Med. Biol., 27–33, 1999.34. The bionic man: restoring mobility, Science, 295, 2002.35. Humayun, M.S., de Juan, E., Jr., Weiland, J. et al., Pattern electrical stimulation

of the human retina, Vision Res., 39, 2569–2576, 1999.



http://hstat.nlm.nih.gov/

36. Liu, W., Humayun, M., de Juan, E., Jr. et al., Retinal prosthesis to benefit thevisual impaired, in Intelligent System and Techniques in Rehabilitation Engineer-ing, Teodorescu, N., Ed., CRC Press, Boca Raton, FL, 2000, chpt. 2.

37. Wyatt, J. and Rizzo, J., Ocular implants for the blind, IEEE Spectrum, 47–53,1996.

38. Chow, A. et al., The subretinal microphotodiode array retinal prosthesis,Ophthalmic Res., 195–198, 1998.

39. Zrenner, E., Stett, A. et al., Can subretinal microphotodiodes successfullyreplace degenerated photoreceptors?, Vision Res., 2555–2567, 1999.

40. Rizzo, J.F., Loewenstein, J., and Wyatt, J., Retinal degenerative diseases andexperimental theory, in Development of an Epiretinal Electronic Visual Prosthesis:The Harvard-Medical Massachusetts Institute of Technology Research Program, Klu-wer Academic, Norwell, MA, 1999, pp. 463–470.

41. Yagi, T., Ito, Y., Kanda, H., Tanaka, S., Watanabe, M., and Uchikawa, Y., Hybridretinal implant: fusion of engineering and neuroscience, in Proc. 1999 IEEEInt. Conf. Systems, Man, Cybernetics, 4, 382–385, 1999.

42. Chow, A.Y. and Chow V.Y., Subretinal electrical stimulation of the rabbitretina, Neurosci. Lett., 225, 13–16, 1997.

43. Humayun, M.S., de Juan, E., Jr., Weiland, J.D., Dagnelie, G., Katona, S., Green-berg, R., and Suzuki, S., Pattern electrical stimulation of the human retina,Vision Res., 39, 2569–2576, 1999.

44. Majji, A.B., Humayun, M.S., Weiland, J.D., Suzuki, S., D’Anna, S.A., and deJuan, E., Jr., Long-term histological and electrophysiological results of aninactive epi-retinal electrode array implantation in dogs, Invest. Ophthalmol.Visual Sci., 40(9), 2073–2081, 1999.

45. Decemberlercq, M., Clement, F., Schubert, M., Harb, A., and Dutoit, M., De-sign and optimization of high-voltage CMOS devices compatible with a stan-dard 5V CMOS technology, in Proc. of the IEEE 1993 Custom Integrated CircuitsConference, 1993, pp. 24.6.1–24.6.4.

46. Ballan, H. and Declercq, M., High Voltage Devices and Circuits in StandardCMOS Technologies, Kluwer Academic, Dordrecht/Norwell, MA, 1999.

47. Allen, P.E. and Holberg, D.R., CMOS Analog Circuit Design, Holt, Rinehart, &Winston, New York, 1987.



chapter seven

Silicon microelectrodes for extracellular recording

Jamille F. Hetke and David J. Anderson

Contents

7.1 Introduction7.2 Extracellular recording7.3 Multichannel recording7.4 Photoengraved microelectrodes

7.4.1 Early attempts7.4.2 Recent developments

7.4.2.1 The Utah Electrode Array7.4.2.2 SOI-based fabrication7.4.2.3 Polyimide electrodes7.4.2.4 The Michigan probe

7.5 The Michigan probe7.5.1 Fabrication7.5.2 Design considerations

7.5.2.1 Shank length7.5.2.2 Shank width7.5.2.3 Substrate thickness7.5.2.4 Site spacing7.5.2.5 Site area

7.5.3 Three-dimensional arrays7.5.4 Acute probes7.5.5 Chronic recordings7.5.6 Example applications

7.5.6.1 Mapping using multi-unit recordings7.5.6.2 Current-source density analysis using

field recordings



7.5.6.3 Chronic recording from behaving animals7.6 Future directionsAcknowledgmentsReferences

7.1 IntroductionWhen envisioning a neural prosthesis, the image that emerges is of electricalstimulation through one or several implanted electrodes to restore lost bodyfunction. While stimulation is the basis for restoration of sensory and motorfunction, recording is also essential in the development and employment ofa prosthesis. First, extensive physiological recording research must be per-formed to investigate and perhaps map an area into which a stimulationdevice will be implanted. The foundation on which a cochlear implant per-forms, for example, is the tonotopic organization of the cochlea. This map,studied and verified in part using extracellular recording, became the basisfor the design of the stimulation prosthesis. Similarly, maps have been iden-tified in primary visual cortex which encode features such as orientation,direction, and color.1–3 Once an implantation target has been chosen, extra-cellular recordings are often used to validate efficacy and optimize the designof implanted stimulation electrodes and protocols. In addition to its use asa basic research tool, the recording device is also an essential component ofclosed-loop prostheses currently being developed for spinal cord injuries. Inthese devices, implanted recording electrodes record biological control sig-nals to drive implanted stimulating electrodes. An example of a closed-loopprosthesis currently under development is a motor prosthesis in which arecording array, implanted in the motor cortex, sends control signals to astimulation device that controls a robotic arm.4–7 Indeed, extracellular record-ing electrodes play a significant role in the world of neural prostheses. Thischapter will begin with a history of recording microelectrodes fabricatedusing thin-film techniques and will describe several of the current researchprojects aimed at developing these electrodes. It will then focus on siliconmicroprobes being developed at the University of Michigan and their design,fabrication and use.

7.2 Extracellular recordingExtracellular recording is one of the most widely used techniques for study-ing the nervous system at the cellular level. When a neuron receives appro-priate stimuli from other cells, its membrane depolarizes and causes ioniccurrents to flow in the surrounding cytoplasm. The voltage drop associatedwith this extracellular current, or action potential, can be measured if a suitableelectrode is located near the active neuron. An extracellular action potentialis typically about 50 to 500 microvolts (

µV) in amplitude, with a frequencycontent from 100 Hz to about 10 kHz. In order to record these neural signals,



the electrode must pass through the extracellular space and approach theactive neuron without damaging it or other cells interacting with it. For thisreason, it is critical that the recording electrode be as small and noninvasiveas possible.

One type of recording electrode is the glass micropipette. These devicesare formed by heating and pulling a 1- to 2-mm diameter glass capillary intotwo pieces. Commercial pipette pullers permit control of the temperatureand force with which the capillary is pulled and therefore control the taperof the resulting tip down to 0.1

µm. Beveling of the tip can also be performedto even more precisely define the tip diameter and impedance. The pulledpipette is filled with an electrolyte solution such as KCl to form a conductivelink to the tissue, and a large-area reversible electrode, inserted in the solu-tion from the top of the pipette, is used to couple to the external world. Theelectrical properties of glass electrodes have been reviewed by Schanne.8 Ingeneral, the equivalent circuit for such a device is dominated by the seriesresistance of the fluid-filled tip and the shunt capacitance of the pipette wall.This forms a low-pass filter which measures DC and low-frequency poten-tials well but rarely permits a frequency response above 1 kHz. Glassmicropipettes are therefore typically used for intracellular recording.

The preferred method for detecting action potentials extracellularly iswith a metal microelectrode. Traditionally, these electrodes have been formedby electrolytically sharpening9,10 or mechanically beveling11,12 a small-diam-eter metal wire (e.g., tungsten, stainless steel, platinum) to a fine tip (<1

µm)and then insulating it, leaving only the tip exposed. Alternatively, microwirescan be formed from preinsulated fine wires that have been cut to expose thecross-sectional area at the end of the wire. The electrical properties of metalmicroelectrodes have been detailed by Robinson.13 Basically, metal micro-electrodes function by forming a capacitive interface to the aqueous tissuemedium, allowing the detection of changes in the potential field created bythe extracellular currents. The electrodes described in the remainder of thechapter are of the metal type.

7.3 Multichannel recordingWhile the recordings from a single electrode site can reveal characteristicsof one or a few cells, they cannot give information about how networks ofcells work together to process information. This requires the use of arraysof microelectrodes to study temporal and spatial relationships betweengroups of neurons. Although only single-channel separation methods existedin the 1960s, the use of such techniques was not disseminated.14 The math-ematical tools for analysis of multichannel records were being developed byGerstein and his colleagues15–17 well before the recording methods wereavailable. These studies revealed the importance of observing the activityand interaction of many neurons simultaneously and led to numerousattempts to construct arrays of microelectrodes by mounting wire electrodeson a common structure. These types of electrodes provided a multiplication



of simultaneous data collection channels and the potential for detectingrelationships among cells. Many reviews exist for this technology and thephilosophy driving it.18 From a signal processing point of view, the separa-tion of neural events recorded on one of the channels depended on variousmethods of waveform discrimination, a method that was well developed bythe early 1980s.19 An important advance in multichannel recording occurredwhen multielectrodes were fabricated with closely spaced sites that hadoverlapping recording volumes. This allowed neurons to be separated byspatial distribution as well as waveform. The stereotrode, introduced byMcNaughton et al.,20 was a close-spaced, two-site wire electrode that couldaccomplish discrimination based on spatial distribution of signals. Drake21

showed the greater discrimination power of a larger number of sites placedclose together on silicon substrates, the principle being that more sitesremove spatial ambiguity and provide greater noise immunity by averagingmultiple sources. Later, a four-wire device was introduced as the tetrode byMcNaughton’s group.22,23 Detailed refinement of the method with quantita-tive evaluation came from Gray.24 A more general approach to electrodeshaving correlated activity was developed by Gozani.25

Various methods of electrode fabrication, including gluing the shanksof metal wire electrodes together, welding glass pipettes together, and depos-iting thin films on silicon substrates, have been used to achieve multichannelrecordings. Microwire electrode arrays are still used extensively today forboth acute and chronic extracellular recording26–28 and can be obtained com-mercially with a variety of different metals and insulators, varying arrayconfigurations, and even with independently positionable electrodes. Ven-dors who offer these arrays and associated instrumentation include FHC,Inc. (www.fh-co.com), NB Labs (www.nblabslarry.com), Thomas Recording(www.thomasrecording.de), and Alpha Omega Engineering (www.alpha-omega-eng.com).

7.4 Photoengraved microelectrodesAlthough arrays of bundled metal microelectrodes have proven to be veryuseful for studying neural circuits, they do have several disadvantages. First,their added volume introduces more tissue damage than does a single con-ductor electrode. In addition, the exact geometrical configuration of suchhand-built arrays is not reproducible. The ability to increase the number ofelectrode recording sites without increasing the volume of the array is anattractive one and spurred the onset of a number of attempts to create amicroelectrode array using high-precision photolithographic techniquesemployed in the microelectronics industry. Using these methods, a singlerecording site can be made about as small as the tip of the finest wireelectrode. However, the microelectrode shank, the portion that supports therecording sites and displaces the tissue, can carry multiple recording sitesand can be at least as small as a single-wire electrode. In general, thesetechniques are the same as those used to create integrated circuits and



http://www.fh-co.com/

http://www.nblabslarry.com/

http://www.thomasrecording.de/

http://www.alphaomega-eng.com/

http://www.alphaomega-eng.com/

therefore utilize similar substrate, conductor, and insulating materials. Fab-rication typically starts on a wafer substrate and the electrode features areadded using a number of photolithographically patterned thin-film layersthat are defined by etching. These methods are attractive because they resultin highly reproducible, batch-processed devices that have features definedto within less than ±1

µm. In addition, many of the techniques are compatiblewith the inclusion of on-chip circuitry. The latter is an important character-istic because, as the density of sites increases, so does the number of inter-connect leads and packaging complexity. In fact, multiplexing becomesimperative as the lead count approaches 32 or more, especially if the elec-trode is to be chronically implanted. In addition, on-chip buffering andamplification can reduce crosstalk, noise coupling and signal attenuation,improving the overall signal-to-noise ratio.

7.4.1 Early attempts

The exploration of photoengraved microelectrodes began in the mid-1960swith efforts at Stanford University.29,30 This silicon substrate structure con-sisted of an array of gold electrodes and conductors insulated by a thin layerof silicon dioxide (Figure 7.1). These probes were capable of recording single-

Figure 7.1 An early photoengraved, multichannel recording electrode. This devicewas based on a silicon substrate and had gold electrodes and conductors insulatedwith silicon dioxide. (From Wise, K.D. et al., IEEE Trans. Biomed. Eng., 17, 238, 1970.With permission.)



unit activity from rat cortex31 and paved the way for future developments.Since that time, a number of efforts have been directed toward the develop-ment of batch-fabricated, photolithographically defined microelectrodearrays. They have differed in their substrate, interconnect, and dielectricmaterials and in methods used for shaping the device. Silicon, tungsten,molybdenum, glass, and polyimide have been among the substrate materialsthat have been explored. Dielectrics that have been investigated includepolyimide, silicon oxide, silicon nitride, and glass. A variety of materialsincluding gold, platinum, tungsten, tantalum, and nickel have been used toform electrode sites. While many of these early attempts resulted in probesthat were capable of recording extracellular activity, they were often low inyield and quality due to difficult fabrication sequences and imprecise meth-ods for shaping of the substrate. Details on these early electrodes can befound in References 32 to 46.

7.4.2 Recent developments

A number of groups continue to develop and improve photoengraved micro-probes using advanced techniques and materials. Recent and ongoing effortsinclude the Utah Electrode Array, arrays based on silicon-on-insulator (SOI)wafer processing, flexible polyimide arrays, and the Michigan probe. For amore inclusive list of recent developments, see References 47 to 65.

7.4.2.1 The Utah Electrode ArrayDr. Richard Normann and his colleagues at the University of Utah havedeveloped a microelectrode array with a high density of penetrating shaftsand referred to as the Utah Electrode Array, or UEA.52–54 Each shaft is 1 to1.5 mm long and projects down from a 0.2-mm-thick glass/silicon compositebase (Figure 7.2). The device is formed from a monocrystalline block ofsilicon using a diamond dicing saw and chemical sharpening. The resultingsilicon shafts are electrically isolated from one another with a glass frit andfrom the surrounding tissue with deposited polyimide or silicon nitride. Thetip-most 50 to 100

µm of each shaft is coated with platinum to form theelectrode site.

Arrays have been fabricated with up to 100 (10

×10) penetrating shaftsthat are spaced on 400

µm centers. Interconnection to the electrode sites isaccomplished by bonding either individual, insulated 25-

µm wires or a mul-tilead polyimide ribbon cable to bond pads on the top of the array. Whilethe shafts are sharp, they are dense and can cause significant tissue dimplingduring normal insertion. A high-velocity insertion technique using a pneu-matic device was developed to alleviate this problem.55 Although the UEAwas originally designed and has been successfully used for acute and chronicrecording in cat cortex,56,57 a modified design has now permitted use in catperipheral nerve.58,59 A variety of acute and chronic versions of the device,along with instrumentation associated with their use, are available commer-cially from Bionic Technologies (www.bionictech.com).



http://www.bionictech.com/

7.4.2.2 SOI-based fabricationAnother relatively new set of research efforts is directed toward developmentof planar neural electrodes fabricated using SOI wafers.60–62 SOI wafers aremanufactured with an oxide layer buried a specified distance below the topsilicon surface (6 to 25

µm for these neural devices). This oxide layer providesan etch-stop that accurately defines the final thickness of the electrodes.Metal conductors and electrode sites are defined photolithographically andare insulated by deposited layers of silicon nitride and silicon dioxide. Deepreactive ion etching (DRIE) is used to etch through the entire thickness ofthe device from the front, stopping on the buried oxide layer. The backsideof the wafer is then patterned and etched through using DRIE to the otherside of the buried oxide layer. The electrodes are finally released by etchingthe buried oxide in buffered HF.

The final thickness of the shanks of these devices is determined by thedepth of the buried oxide layer. Shanks from 6 to 25

µm have been fabricated.Bond pad regions are typically left at full wafer thickness (>500

µm). At thetime of this writing, single and multiple shank acute designs have been fab-ricated with up to 32 sites and have been used successfully for acute record-ings. In anticipation of chronic use, one group has added a process step toinclude surface topology on the devices in an effort to improve mechanicalanchoring in the tissue.60 In addition to being compatible with the inclusionof on-chip circuitry, this type of fabrication offers the advantage that all of theprocess steps are compatible with current foundry techniques.

7.4.2.3 Polyimide electrodesAn alternative approach that is being investigated for chronic multichannelrecording is flexible substrate electrodes. While a flexible substrate may be

‘

Figure 7.2 The Utah Electrode Array (UEA). This 10

×10, silicon-based array hasplatinum electrode sites on the tip of each shaft. The shafts are spaced on 400-

µmcenters. (From Nordhausen, C.T. et al., Brain Res., 726, 129, 1996. With permission.)



difficult to insert into the brain without an insertion aid, it has been hypoth-esized that once it is successfully implanted its long-term performance maybe superior to silicon because its elastic modulus more closely mimics thatof the brain and it may therefore reduce trauma due to micromotion.Advancements in the material polyimide, originally investigated in the early1980s as an electrode substrate,39,40,42–44 have caused it to be recently revisitedby several groups.63–65 Basically, these planar structures consist of metal tracessandwiched between upper and lower layers of polyimide. The processbegins with a silicon wafer that acts as a carrier throughout the process.Depending on how the devices are to be released from the wafer, a sacrificiallayer may first be deposited on the wafer. Polyimide is then spun on to thedesired thickness. If the polyimide is photodefinable it is next exposed anddeveloped to define the base of the final device. If the polyimide is notphotodefinable, RIE can be used for etching. Metal is deposited, patterned,and etched to form the interconnects. The top layer of polyimide is spun on,exposed, and developed to define the openings to the underlying metalwhich will become the recording sites and bond pads. Finally, the devicesare released from the wafer either by dissolving the sacrificial layer in a wetetch65 or by peeling them off with tweezers.63

Although it has not yet been attempted, these devices may be compatiblewith the inclusion of on-chip circuitry. For the interim, the Fraunhofer Insti-tute has developed a novel technique for high-density interconnection ofhybrid integrated circuit (IC) chips to polyimide devices.66 In addition, unlikethe two silicon-based devices described above, an integrated flexible cablecan be extended off the electrode for chronic connection to the external worldvia a percutaneous connector. The final devices have thicknesses less than20

µm. The Fraunhofer devices are designed for chronic interfacing withperipheral nerves. Rousche and his colleagues65 are working toward achronic intracortical array and have made three-dimensional structures byfolding the two-dimensional devices into the appropriate shape (Figure 7.3).Both types of arrays have successfully obtained chronic recordings. Devicesfabricated by Rousche et al. have “wells” included in the substrate that aredesigned to be seeded with bioactive materials to improve biological accep-tance of the device.

7.4.2.4 The Michigan probeOur group at the University of Michigan has developed a planar microelec-trode array useful for both recording and stimulation. The probes, for whichthe basic structure is shown in Figure 7.4, are based on a silicon substrate,the thickness and shape of which are precisely defined using boron etch-stop micromachining.67,68 The substrate supports an array of conductors thatare insulated by thin-film dielectrics. Openings through the upper dielectricsare inlaid with metal to form the electrode sites for contact with the tissue,and the bond pads for connection to the external world. The Michigan probeis the focus of the remainder of this chapter.



7.5 The Michigan probe7.5.1 Fabrication

While the Michigan probe process is compatible with the inclusion of on-chip CMOS (complementary metal-oxide semiconductor) circuitry for signalconditioning and multiplexing and neural recordings have been made using

Figure 7.3 A flexible polyimide electrode array. This 12-channel device, developedat Arizona State University, has been folded to achieve a three-dimensional structure.The shanks are spaced on 100-

µm centers. (From Rousche, P.J. et al., IEEE Trans.Biomed. Eng., 48, 361, 2000. With permission.)

Figure 7.4 Basic structure of the Michigan silicon probe.

Output leads

CMOS circuitry(optional)

Electrode sites

Conductors



these active arrays,69 this chapter centers on the passive electrode array,which does not include circuitry. The Michigan passive probe process, out-lined in Figure 7.5, starts with a silicon wafer of standard orientation <100>and thickness (500

µm). The wafer is thermally oxidized and this oxide isthen selectively removed using photolithography to define the shape of theprobe. Boron is diffused into the silicon in these open areas. Because theboron-doped silicon acts as an etch stop during the final release of the probes,the final thickness of the devices is defined by the depth of this diffusion. A

Figure 7.5 Fabrication sequence for the Michigan passive probe (not to scale). Fabri-cation requires only single-sided processing and uses eight masks.

Grow and pattern thermal oxide maskBoron diffuse to form substrate

Deposit lower dielectricsDeposit and pattern conductors

Deposit upper dielectrics

Open contact viasDeposit and liftoff sites and bond pads

Etch field dielectricsRelease probes in EDP

25µm



standard deep diffusion results in a final device thickness of 15

µm. Afterdiffusion, the masking oxide is stripped and a tri-layer stack of dielectrics(silicon dioxide, silicon nitride, and another layer of silicon dioxide) is depos-ited using low-pressure chemical vapor deposition (LPCVD). These dielec-trics insulate the lower substrate from the thin-film interconnects that willbe deposited in the next step. They are put down in this configuration andin thicknesses that result in a neutral mechanical stress condition (3000, 1500,3000 Å) so that a flat device results in the end.

Next, the thin-film interconnect material that will lead from the electrodesites to the bond pads is deposited, patterned using photolithography, andetched using reactive ion etching (RIE) to form the multiple leads. Thematerial typically used for recording interconnects is 8000 Å of phosphorus-doped polysilicon. The traces are then insulated from above using the sametri-layer stack of dielectrics used for lower insulation. Contact vias are nextopened using a combination of wet (HF) and dry (RIE) etching to exposethe interconnect traces for metallization which will form the electrode sitesand bond pads. These sites and bond pads are formed using sputtering andthen a lift-off process that removes the deposited metal from the wafereverywhere except in the electrode site and bonding pad areas. Typically,the electrode sites are made from iridium with a thin underlayer of titaniumfor adhesion, and the bond pads are made from gold with a chromiumadhesion layer. Other site materials, including gold, titanium nitride, carbon,and platinum have been used depending on the intended application. Itshould be noted here that if iridium is used for the sites, the sites can beelectrochemically activated and are suitable for stimulation as well as record-ing.70–72 The next step in the process is to remove the dielectrics surroundingthe probe shapes, which is accomplished using RIE.

The final process steps are performed to release the probes from thewafer. First the wafer is thinned from the backside using an unmaskedetch in a hydrofluoric-nitric acid mixture. The thinned wafer is finallyplaced in an anisotropic silicon etch composed of ethylenediamine, pyro-catechol, and water (EDP). EDP etches away all of the undoped silicon andstops on the boron-doped areas. It does not attack any of the other exposedmaterials used in the probe process. Once they are removed from EDP, theprobes are rinsed clean in solvents and deionized water and are ready forsorting and packaging.

One additional masking step can be added to the above process toallow the formation of a sharper probe tip, a thinner substrate, and/or anintegrated ribbon cable for chronic use.73 This extra step, a shallow borondiffusion, is performed just after the deep diffusion. The rest of the process(i.e., deposition and patterning of the leads, dielectrics, and sites) is carriedout as described above. While the deep diffusion results in a final devicethickness of about 15

µm, the shallow-diffused device thickness is onlyabout 5

µm. The probe substrate is therefore thick enough to penetrate thepia of most animals; the shallow-diffused region, if used to form an inte-grated cable, is thin and flexible and minimizes tethering forces on an



implanted probe. Probe tips formed using shallow diffusion have a radiusof about 1

µm and enter the tissue with minimal dimpling. In addition,these sharp tips may be useful for penetrating tougher structures such asnerve and spinal cord.

The Michigan passive probe process with shallow boron diffusion useseight masks. It is capable of yields in excess of 90%. Typically, 10 to 15different designs are included on a given mask set and, depending on devicesizes, a 4-inch silicon wafer can produce well over 1000 probes. Passivedevices with lengths of 1.5 mm to 5 cm and widths as narrow as 5

µm havebeen fabricated with as many as 96 sites.

7.5.2 Design considerations

One of the attractive features of the Michigan and other planar photoen-graved probes is the ability to customize design for specific experiments.The substrate can have any two-dimensional shape with single or multipleshanks, electrode sites can be of any surface area and can be placedanywhere along the shank(s) at any spacing, tips can be made very sharpor blunt, and features such as holes and barbs can be included for specialapplications such as sieve electrodes.74–76 The NIH NCRR-sponsored Uni-versity of Michigan Center for Neural Communication Technology(CNCT) has been providing probes to investigators since 1994. At the timeof this writing, over 150 designs have been fabricated through the CNCT,a subset of which comprise a catalog of basic designs. A number ofscientific papers and presentations, listed at www.engin.umich.edu/facil-ity/cnct/papers.html, have resulted from use of these probes. The designfreedom offered by this technology makes the devices attractive to research-ers in a broad range of applications; however, there are some practical andprocess-enforced limitations on probe features that must be consideredwhen selecting or designing a probe.

7.5.2.1 Shank lengthWhen selecting or designing a probe, a first characteristic an investigatortypically considers is shank length. This is the length of the device that willbe inserted into tissue or the maximum depth of the intended target. Whileshank length can be considered a process-limited feature due to the maxi-mum limit imposed by the size of the wafer, it is actually more practicallylimited as defined by stiffness and strength. Typical 15-

µm-thick, 100-

µm-wide probe shanks longer than about 6 mm may bend during insertion,causing the target to be missed. As described by Najafi and Hetke,77 whenchoosing a shank length one must consider the buckling load, which definesthe stiffness of the probe as it is pressed against the tissue, and the maximumstress at the bottom of the probe shank, which defines the strength of theprobe. The buckling load is proportional to t3W/L,2 and the maximum stressis proportional to t/L.2 Therefore, to design a probe that is stiff and strongwith the given thickness t = 15

µm, the substrate width (W) should be



http://www.engin.umich.edu/

http://www.engin.umich.edu/

increased, and the length (L) should be the minimum necessary to reach theintended target.

Insertion properties can also be improved by increasing the sharpnessof the tip.78 As described above, this can be achieved by using a shallowboron diffusion at the probe tip which results in a structure that is no morethan 1 to 2

µm in diameter. In addition to the shallow diffusion, a verysharp taper angle (<10 degrees) can be used to improve penetration char-acteristics. Figure 7.6 shows a probe tip that is shallow diffused and has ataper angle of 18 degrees that has successfully penetrated sciatic and audi-tory nerves in cat, structures which are not penetrable with standard deepdiffused tips.

7.5.2.2 Shank widthTaking into consideration the above discussion on shank length and therelationship of shank width to the strength and stiffness of long shanks, oneshould still try to minimize the width of the shank to make the device asnoninvasive as possible. Shank width is a process enforced limitation thatis controlled by the diffusion mask and lateral diffusion. As the width of thediffusion opening decreases, lateral diffusion of boron underneath the maskeventually limits the minimum achieveable shank width for a given substratethickness.79 Using an EDP etch alone, the minimum shank width achievablefor a standard deep-diffused 15-

µm-thick shank is about 15

µm because, asthe mask width decreases, so does the thickness. Narrow “scaled” probescan be realized, however, using shallow-diffusion and/or an additionalmasking and etching step. To do this, deep RIE is used to etch into the siliconsubstrate and trim off the lateral diffusion, resulting in a narrower devicethan would be realizable using the EDP etch alone. Scaled probes have beenfabricated with shanks as narrow as 5

µm. While 1.5-mm-long, 5-

µm-thickscaled shanks have successfully penetrated the guinea pig pia mater andrecorded from cortex, the mechanical practicality of these tiny devices hasyet to be proven.

Figure 7.6 A sharp probe tip formed using a combination of shallow and deep borondiffusions. Probes with these sharp tips have been used to penetrate tough structuressuch as peripheral nerve and spinal cord.



Another issue related to shank width is the interconnect lead width andspacing. Although submicron feature sizes are standard in industry, one mustconsider electrical crosstalk between very close leads. When more electrodesare to be accommodated on a narrower shank, the width of interconnectlines and the spacing between the lines must be reduced, resulting in anincreased coupling capacitance. As described by Najafi et al.,79 for probeswith line widths and spacing as small as 1

µm, the crosstalk approaches 1%.This is acceptable for recording devices, as effects will be negligible com-pared to background noise. Even with feature sizes as small as 0.25

µm, thecrosstalk is still less than 4%.

7.5.2.3 Substrate thicknessSubstrate thickness is a feature that is of importance when it comes toinsertion into tough tissue or into deep structures. The standard Michiganprobe process is capable of producing probes 5 to 15

µm in thickness basedon the time and temperature of the boron diffusion and hence the depth ofthe etch-stop. Probes 15

µm thick and 5 mm long and tapered from 15 to120 µm in width are capable of penetrating the pia with minimal bendingin most preparations. If the device is much longer than this, however, thebuckling force is lower. If a very long shank is absolutely required by theapplication, a stronger, stiffer device can be achieved by forming a “box-beam” substrate. This type of device uses the same fabrication process asthe Michigan chemical delivery probe80 to form an open channel within thesilicon substrate (Figure 7.7). This channel results in a box-beam structurethat is inherently stiffer and stronger than a normal beam because the secondmoment of its cross-section is larger. This structure is being investigated forits efficacy in penetrating deeper and/or tougher structures, including theinferior colliculus, spinal cord, and cochlear nucleus in cats.

Figure 7.7 A probe with three channels, spaced 20 µm apart, buried within its sub-strate. While these probes are being developed for chemical delivery, they are alsouseful for applications where a stiffer device is necessary. (From Chen, J. et al., IEEETrans. Biomed. Eng., 44, 760, 1997. With permission.)



7.5.2.4 Site spacingAnother characteristic one considers when choosing a probe design is thecenter-to-center spacing between the recording sites. This feature is basicallyunlimited; the spacing between sites can be as small as a few microns to aslarge as several hundred microns or more. The choice will depend on thelocation of intended use as well as the type of data to be collected. Forexample, for purposes of spike sorting, signal improvement, and/or cellimaging, correlated signals are desired on more than one site. A general ruleof thumb for these types of recordings is that the sites should be spaced at50 µm or less.21 Close-spaced sites can be in the form of a linear array, or ina tetrode configuration as shown in Figure 7.8. Sites that are spaced 100 µmapart will show some correlated as well as uncorrelated activity. When thesites are spaced at 200 µm or more, little or no coherency occurs betweensites.21 Arrays with this larger spacing, for example, can be used to investi-gate the layered organizations of cortical tissue.

7.5.2.5 Site areaThe choice of site area for recording is an electrical consideration as it isrelated to thermal noise and the ability to record low-level signals and henceinfluences the achievable signal-to-noise ratio. There is a trade-off whenchoosing a site area. Although perhaps more selective in the signals it detects,a smaller site imposes a higher impedance, causing signal attenuation andadded noise. In contrast, a larger site offers a lower impedance at the expenseof being less selective and therefore picks up signals from a larger populationof cells. While site areas on Michigan probes ranging from 70 to 4000 µm2

Figure 7.8 A two-shank tetrode array; sites are spaced at 25 µm on the diagonal, andshanks are spaced at 150 µm.



have successfully recorded neural spikes, the current typical recording siteon a Michigan probe is just under 200 µm2. The average 1-kHz impedanceon a 200-µm2 iridium site is around 2 MΩ and, with appropriate amplifiers,the baseline noise level is about 15 µVRMS. This site area has evolved overmany years as a “best” choice for maximizing signal-to-noise ratio and, infact, is still evolving. It is likely that a single best choice for all brain regionsdoes not exist due to variations in cell size and density. A systematic studyis currently underway to investigate the recording quality of different siteareas in a variety of brain structures.

If the electrode site is to be used for stimulation, a larger site size maybe required to safely deliver the charge required for the application. Inaddition, the charge capacity of the site can be increased by an order ofmagnitude by forming iridium oxide through electrochemical activation.70,72

This procedure involves cycling the electrode potential between positive andnegative limits, which results in a porous, hydrous, multilayer oxide filmthat has a high charge capacity and is exceptionally resistant to dissolutionand corrosion during stimulation.

7.5.3 Three-dimensional arrays

The Michigan probe process results in devices that are planar and twodimensional. Because neural systems are three dimensional, however, inorder to fully instrument the target tissue volume it is important to be ableto realize three-dimensional arrays. Such three-dimensional structures canbe constructed from the two-dimensional components using microassemblytechniques.81,82 The procedure is based on inserting multiple two-dimen-sional probes into a silicon micromachined platform that is intended to siton the cortical surface. Silicon spacers are inserted over the probes to holdthem parallel to one another. Gold tab bond pads on the probes are bent atright angles to form a flat contact with mating pads on the platform. Ultra-sonic bonding of these pad pairs is performed to establish electrical contactbetween the probes and the platform. Exposed connections are then pottedin silicone rubber. An integrated cable on the platform carries the signals toa percutaneous connector.

Although these three-dimensional arrays are dense, they occupy lessthan 1% of the tissue volume into which they are inserted. Each shank ofthe array is only 15 µm thick. Recording arrays having up to 8×16 shankson 200-µm centers have been fabricated, constructed, and used to recordacute single units from guinea pig cortex. Following the methods of the Utahgroup, a dynamic insertion tool is used to insert the arrays into cortex withminimal traumatic injury. Unlike the UEA, these arrays can have multiplesites along each shank. This makes the need for on-chip electronics obvious;the signal leads must be multiplexed prior to lead transfer between the probeand the platform in order to make the transfer feasible. A device with on-chip CMOS signal processing circuitry has been created and used to recordacute neural activity (Figure 7.9).



7.5.4 Acute probes

Acute recordings are made on anesthetized animals and continue anywherefrom several hours to several days. These experiments are less complex thanchronic experiments because the surgery is simpler, the animal is not awakeduring recordings, and the electrode assembly is not subjected to the corro-sive physiological environment for an extended time. The acute Michiganprobe assembly is shown in Figure 7.10. The probe is electrically connectedto a printed circuit board using ultrasonic bonding, and the exposed con-nections are insulated with silicone rubber. This package is easy to handle

Figure 7.9 A three-dimensional silicon array that has been microassembled fromplanar components. This device has on-chip CMOS preamps and buffers. (From Bai,Q. et al., IEEE Trans. Biomed. Eng., 47, 281, 2000. With permission.)

Figure 7.10 Acute 16-channel Michigan probe assembly. The PC board has pins con-figured to plug into a standard IC socket.



and use in most acute preparations. The pins on the circuit board are con-figured to plug into a standard IC socket which makes the package easilyadaptable to most existing systems. In addition, several commercial vendorshave developed headstage amplifiers that mate directly to Michigan probes.At the time of this writing, these vendors include Plexon, Inc. (www.plex-oninc.com), Neuralynx, Inc. (www.neuralynx.com), and Tucker-Davis Tech-nologies (www.tdt.com).

These acute headstages are intended to be mounted on a micromanip-ulator to permit the attached probe assembly to be lowered precisely intothe brain. Prior to insertion, the dura must be pierced to permit probepenetration. Buckling of the shank can be minimized by inserting the probeshank(s) at an angle normal to the brain’s surface. Once the device is loweredinto the desired area, warm agar solution may be used over the exposedtissue to prevent dessication and minimize pulsation artifact. Because acuteprobes are not typically fixed in place with cement or acrylic, they can oftenbe recycled several times if they are rinsed immediately upon withdrawalfrom the tissue with deionized water.

7.5.5 Chronic recordings

Chronic recordings are required in a variety of studies including those relatedto learning, adaptation, and plasticity. Stable chronic recordings for periodsextending over years are critical if an electrode is to be used for studying orimplementing a neural prosthesis. These recordings are more difficult toobtain than acute recordings not only because the animals are typicallyawake and sometimes behaving, but also because the electrode/tissue inter-face must remain stable over time. The natural response of the tissue to theforeign electrode body is to encapsulate it in a cellular sheath, which isolatesthe recording sites from the brain,83 causing an eventual degradation ordisappearance of recordings. All electrode types (i.e., wire, silicon, polyim-ide) are plagued with this problem to a certain degree and there are currentlymany efforts focused on improving the tissue interface.

Although telemetry systems are under development that will eliminateoutput leads and hence improve positional stability and perhaps the tissueinterface, most systems in use today are tethered. These assemblies can beclassified as fixed, movable, or floating. A fixed electrode is inserted into thebrain and the proximal end is permanently attached to the skull or chamberabove. Microwire arrays, arrays of individually insulated wires bundledtogether,27 can be considered to be of the fixed type. While these devices mayfunction well for recording from deep structures and from the cortex of smallanimals for periods extending over a year, larger animals and humans exhibitgreater brain movement with respect to the skull which may lead toincreased relative movement of the electrode and more tissue damage.Avoidance of this type of damage is especially critical for very long-termimplants such as will be required for a prosthesis. The movable electrode, asits name implies, is one that is moved periodically to record from similar or



http://www.plexoninc.com/

http://www.plexoninc.com/

http://www.neuralynx.com/

http://www.tdt.com/

even different brain regions over time.84–87 Although these electrodes can bequite useful for certain experiments and in fact may alleviate some of theproblems associated with electrode failure due to encapsulation, they arenot suitable for studies that require chronic recordings from the same cellsover time and are probably not the choice for use in a prosthesis. For long-term chronic use, at least until telemetry systems can be fully implemented,the floating electrode is most likely the device of choice. In this configuration,a flexible interconnect between the implanted electrode and the percutane-ous connector permits the electrode to move with the brain’s natural pulsa-tions, minimizing damage caused by relative movement. A classic exampleof a floating electrode is the “hatpin” electrode developed by Schmidt et al.,88

which remained functional for 1144 days in monkey cortex.While there are examples of applications where the Michigan probe has

been used in short-term chronic experiments in fixed and movable configu-rations,89–91 the chronic assembly that is being developed for long-term useis of the floating type. The flexible interconnect to the probe is achieved usingthe integrated, multilead silicon ribbon cable described above. These 5-µm-thick cables are extremely flexible; even with multiple leads they are roughly100 times more flexible than a single 25-µm-diameter gold wire.73 Becausethe probe and ribbon cable are monolithic, no further electrical connectionis required between the two. The leads are brought out to the external worldthrough a percutaneous connector. An intermediate PC board serves to trans-fer the ultrasonically bonded leads from the probe to the pins on the con-nector (Figure 7.11). Currently, 16- and 32-channel versions of this assemblyhave been developed. Recording periods of over one year have beenachieved with this device.73 The vendors mentioned above are developing

Figure 7.11 Chronic 16-channel Michigan probe assembly. An integrated silicon cableforms the interconnect between the probe and percutaneous connector.



miniature headstages that mate to these assemblies, which will be suitablefor use on behaving animals.

Appropriate surgical techniques are the first step in the realization of asuccessful chronic implant. The probe assembly should first be sterilizedusing ethylene oxide. As with the acute probes, the dura must be removedat the insertion point to permit penetration. The percutaneous connector isthen mounted on the skull using dental acrylic. The probe is next insertedeither by hand by grasping the silicon bead at the probe/cable junction withforceps or with a vacuum-pick mounted on a micromanipulator. Quick-curesilicone rubber can then be placed over the implanted electrode to providesupport for it and protection for the exiting cable. Dental acrylic is finallyused to seal the entire area.

7.5.6 Example applications

The Michigan probes have been utilized in a variety of brain regions forextracellular recording of unit and field activity. Several applications, acuteand chronic, will be highlighted here.

7.5.6.1 Mapping using multi-unit recordingsDr. John Middlebrooks and his colleagues at the University of Michiganare using acute probes to map “cortical images” of activation in guineapig auditory cortex in response to stimulation through a cochlearimplant.92 The goal of these studies is to identify electrode configurationsthat optimize information transmission from the cochlear implant to thecortex and to use the results to improve the design of speech processingstrategies. In these experiments, a silicon probe was used to record multi-unit auditory cortical activity elicited by cochlear implant stimuli thatvaried in electrode configuration, location of stimulation within thecochlea, and stimulus level. Cochlear electrode configurations that wereinvestigated were monopolar (MP), bipolar (BP+N) with N active elec-trodes between the active and return electrodes, tripolar (TP) with oneactive electrode and two flanking return electrodes, and common ground(CG) with one active electrode and up to five return electrodes. The probedesign that was chosen for these experiments was a 16-channel single-shank probe with 100-µm site spacing. This design permitted simulta-neous recording of the resulting spatio-temporal pattern of neural activityacross the tonotopic axis of the auditory cortex. Figure 7.12 shows anexample of cortical images resulting from various cochlear stimulatingelectrode configurations (rows) and stimulus levels (columns). This figureillustrates several effects. First, it can be seen that the cortical imagesincrease in width in response to increasing current levels. In addition, atthe 7-dB level, the images resulting from MP, BP, and CG stimulationspanned nearly the entire 1.5 mm of the cortical recording array, whilethe image resulting from the TP stimulus remained more restricted evenat the higher current levels.



7.5.6.2 Current-source density analysis using field recordingsIn addition to their usefulness in studying single- and multi-unit extracellu-lar activity, by broadening the bandwidth of external instrumentation theprobes can be used to record field or population potentials. Dr. GyorgyBuzsaki’s group at Rutgers University uses Michigan probes to investigatesimultaneous unit and field activity in hippocampus and neocortex. He haspublished several studies in which he performs current-source density (CSD)analysis to precisely localize current sources and sinks in the awake rat inorder to investigate the cellular-synaptic generation of physiologically rele-vant cortical potentials.93–96 CSD profiles are derived from field recordings.Changes in the sign of the CSD as a function of depth are interpreted as achange in the local divergence of current flow and correspond to the presenceof current sources or sinks.97

In one example of Buzsaki’s CSD work,94 he used Michigan probes tostudy the mechanism of extracellular current generation in the neocortexunderlying sleep spindles and spike-and-wave patterns, two low-frequency(<15 Hz) rhythms associated with the thalamocortical system. These spon-taneous rhythms were compared to evoked potentials in response to stim-ulation of various thalamic nuclei. Field potentials and unit activity were

Figure 7.12 Cortical images, derived from Michigan probe recordings for variouscochlear-stimulating electrode configurations (rows) and stimulus levels (columns)in the guinea pig. The y-axis of each panel represents the silicon probe recording sitelocation relative to the most caudal site, and the x-axis represents the time afterstimulus onset. The gray scale represents normalized spike probability (see originalpublication for color version of figure). (From Bierer, J. A. and Middlebrooks, J. C.,J. Neurophysiol., 87, 478, 2002. With permission.)



recorded simultaneously along trajectories perpendicular to the cortical lay-ers using a 16-site silicon probe with a site spacing of 100 µm. For theseshort-term chronic studies, recordings were obtained in both the anesthe-tized and the awake rat using a movable probe.

In Figure 7.13, voltage vs. depth profiles is superimposed onto CSD mapsof spike-and-wave discharges (HVS) and thalamic evoked responses (VPLi,VL, VPLc) in the awake rat and during Ketamine anesthesia. Thalamic evokedfield responses and spontaneously occurring spike-and-wave dischargesresulted in spatially similar localized sinks and sources in the various corticallayers. The findings from this study indicate that the major extracellular cur-rents underlying sleep spindles, HVSs, and evoked responses result fromactivation of intracortical circuitry rather than from thalamocortical afferentsalone. The multisite recording technique allowed for the continuous, simulta-neous recording of field potentials, current-source density distributions, and

Figure 7.13 Voltage vs. depth profiles, obtained using Michigan probes, superim-posed on CSD maps of spike-and-wave (HVS) and thalamic evoked responses (VPLi,VL, VPLc) in awake and anesthetized rats. Thalamic evoked field responses andspontaneously occurring spike-and-wave discharges resulted in spatially similar lo-calized sinks and sources in the various cortical layers (see original publication forcolor version of figure). (From Kandel, A. and Buzsaki, G., J. Neurosci., 17, 6783, 1997.With permission.)



unit activity. The Michigan probe has become a powerful tool for CSD analysis.Other examples of CSD studies performed using Michigan probes can befound in References 98 to 102.

7.5.6.3 Chronic recording from behaving animalsDr. Daryl Kipke of the University of Michigan is working to characterizethe electrode-tissue interface with the aim of developing techniques tobetter control this interface to improve long-term functionality for even-tual clinical use. Kipke’s group is studying a variety of electrode typesincluding microwires,27 polyimide,65 and the Michigan probe. For thesestudies, the probes are implanted into cerebral cortex of the rat. Long-term recordings (months to over a year) have been obtained in all suc-cessful implants. The longest implant to date has been 15 months, at whichtime the experiment was terminated for complications unrelated to elec-trode stability. Signal-to-noise ratios have remained high throughoutimplant durations, as evidenced in Figure 7.14. This figure shows datataken 382 days after the implant.

7.6 Future directionsSolid-state processing technology has permitted the realization of a numberof novel devices for extracellular multichannel recording. Advantages overconventional metal electrodes include batch fabrication, highly reproduciblegeometrical and electrical characteristics, and small feature sizes. In addition,some of the devices described are compatible with on-chip circuitry for signalconditioning. These devices are being used routinely in acute and short-termchronic experiments for recording single units and extracellular field poten-tials. The future power of these devices, however, will lie in their ability tomake long-term chronic connections with the brain to provide control signals

Figure 7.14 Chronic data from a Michigan probe 382 days post-implant. The probewas placed in barrel cortex of the rat. The left panel shows analog data from onechannel (sample rate, 40 kHz) that has been evoked using whisker stimulation, andthe right panel shows signal-to-noise ratios for all 16 channels of the probe. (Datacourtesy of Rio Vetter and Dr. Daryl Kipke.)



for closed-loop prosthetic systems. There are several areas of researchfocused on improving long-term chronic recordings.

Bioactive coatings hold much promise for improving the electrode-tissueinterface and enhancing chronic recording stability.103–105 Coating the surfaceof the electrode may be advantageous for several reasons. First, the coatingmay act as a mechanical buffer between a relatively rigid electrode substrateand the soft neural tissue. In addition to making the electrode more tissuefriendly, this may improve positional stability. Second, the coating topologycould be made such that it promotes favorable geometrical reactions withcells. Finally, it may be possible to seed the coating with biologically activemolecules such as neurotrophins or living cells to positively influence cellu-lar behavior.

Reducing the number of output leads should also improve chronic life-time by limiting tethering forces on the probe, thereby reducing relativemicromotion, migration, and consequent adverse tissue response. Siliconprobes are being developed with on-chip CMOS circuitry to amplify, multi-plex, and transmit analog neural activity using a total of only three leads forpower and bidirectional data transfer. For future prosthetic systems, how-ever, data and power transfer between the implanted unit and the outsideworld should be achieved without any interconnect leads. Circuits and tech-nologies are being developed for wireless operation and control of implant-able recording106,107 and stimulation108–110 microsystems.

AcknowledgmentsThis work is supported by NIH NINDS (N01-NS7-2364) and NIH NCRR(P41-RR09754).

References1. Hubel, D.H. and Wiesel, T.N., Receptive fields, binocular interaction and

functional architecture in the cat’s visual cortex, J. Physiol., 160, 106, 1962.2. Swindale, N.V., Matsubara, J.A., and Cynader, M.S., Surface organization of

orientation and direction selectivity in cat area 18, J. Neurosci., 7, 1414, 1987.3. Lennie, P., Krauskopf, J., and Sclar, G., Chromatic mechanisms in striate cortex

of macaque, J. Neurosci., 10, 649, 1990.4. Chapin, J.K. et al., Real-time control of a robot arm using simultaneously

recorded neurons in the motor cortex, Nature Neurosci., 2, 664, 1999.5. Isaacs, R.E., Weber, D.J., and Schwartz, A.B., Work toward real-time control

of a cortical neural prosthesis, IEEE Trans. Rehab. Eng., 8, 196, 2000.6. Serruya, M.D. et al., Instant neural control of a movement signal, Nature, 416,

141, 2002.7. Taylor, D.M., Tillery, S.I.H., and Schwartz, A.B., Direct cortical control of three-

dimensional neuroprosthetic devices, Science, 296, 1829, 2002.8. Schanne, O.F. et al., Electrical properties of glass microelectrodes, Proc. IEEE,

56, 1072, 1968.



9. Marg, E. and Adams, J.E., Indwelling multiple microelectrodes in the brain,Electroencephalogr. Clin. Neurophysiol., 23, 277, 1967.

10. Kaltenbach, J.A. and Gerstein, G.L., A rapid method for production of sharptips on preinsulated microwires, J. Neurosci. Meth., 16, 283, 1986.

11. Kruger, J. and Bach, M., Simultaneous recording with 30 microelectrodes inmonkey visual cortex, Exp. Brain Res., 41, 191, 1981.

12. Reitbock, H.J., Fiber microelectrodes for electrophysiological recordings,J. Neurosci. Meth., 8, 249, 1983.

13. Robinson, D.A., The electrical properties of metal microelectrodes, Proc. IEEE,56, 1065, 1968.

14. Abeles, M. and Goldstein, M.H., Multispike train analysis, Proc. IEEE, 65,1977, 762.

15. Gerstein, G.L. and Clark, W.A., Simultaneous studies of firing patterns inseveral neurons, Science, 143, 1325, 1965.

16. Gerstein, G.L. and Perkel, D.H., Simultaneously recorded trains of actionpotentials: analysis and functional interpretation, Science, 164, 828, 1969.

17. Gerstein, G.L. and Perkel, D.H., Mutual temporal relationships among neu-ronal spike trains, Biophys.J., 12, 453, 1972.

18. Eggermont, J.J., Johannesma, P.M., and Aertsen, A.M., Reverse-correlationmethods in auditory research, Q. Rev. Biophys., 16, 341, 1983.

19. Wheeler, B.C. and Heetderks, W.J., A comparison of techniques for classifica-tion of multiple neural signals, IEEE Trans. Biomed. Eng., 29, 752, 1982.

20. McNaughton, B.L., O’Keefe, J., and Barnes, C.A., The stereotrode: a newtechnique for simultaneous isolation of several single units in the centralnervous system from multiple unit records, J. Neurosci. Meth., 8, 391, 1983.

21. Drake, K.L. et al., Performance of planar multisite microprobes in recordingextracellular activity, IEEE Trans. Biomed. Eng., 35, 719, 1988.

22. Wilson, M.A. and McNaughton, B.L., Dynamics of the hippocampal ensemblecode for space, Science, 261, 1055, 1993.

23. O’Keefe, J. and Recce, M.L., Phase relationship between hippocampal placeunits and the EEG theta rhythm, Hippocampus, 3, 317, 1993.

24. Gray, C.M. et al., Tetrodes markedly improve the reliability and yield ofmultiple single-unit isolation from multi-unit recordings in cat striate cortex,J. Neurosci. Methods, 63, 43, 1993.

25. Gozani, S.N. and Miller, J.P., Optimal discrimination and classification ofneuronal action potential waveforms from multiunit, multichannel recordingsusing software-based linear filters, IEEE Trans. Biomed. Eng., 41, 358, 1994.

26. Nicolelis, M.A.L. et al., Reconstructing the engram: simultaneous, multisite,many single neuron recordings, Neuron, 18, 529, 1997.

27. Williams, J.C., Rennaker, R.L., and Kipke, D.R., Long-term neural recordingcharacteristics of wire microelectrode arrays implanted in cerebral cortex,Brain Res. Protocols, 4, 303, 1999.

28. Porada, I. et al., Rabbit and monkey visual cortex: more than a year of re-cording with up to 64 microelectrodes, J. Neurosci. Meth., 95, 13, 2000.

29. Wise, K.D., Angell, J.B., and Starr, A., An integrated-circuit approach to ex-tracellular microelectrodes, IEEE Trans. Biomed. Eng., 17, 238, 1970.

30. Wise, K. and Angell, J.B., A low-capacitance multielectrode probe for use inextracellular neurophysiology, IEEE Trans. Biomed. Eng., 22, 212, 1975.



31. Starr, A., Wise, K.D., and Csongradi, J., An evaluation of photoengravedmicroelectrodes for extracellular single-unit recording, IEEE Trans. Biomed.Eng., 20, 291, 1973.

32. Prohaska, O. et al., A multielectrode for intracortical recordings produced bythin-film technology, Electroencephalogr. Clin. Neurophysiol., 42, 421, 1977.

33. Mercer, H.D. and White, R.L., Photolithographic fabrication and physiologicalperformance of microelectrode arrays for neural stimulation, IEEE Trans.Biomed. Eng., 25, 494, 1978.

34. Pickard, R.S., A review of printed circuit microelectrodes and their produc-tion, J. Neurosci. Meth., 1, 319, 1979.

35. Pochay, P. et al., A multichannel depth probe fabricated using electron beamlithography, IEEE Trans. Biomed. Eng., 26, 199, 1979.

36. May, G.A., Shamma, S.A., and White, R.L., A tantalum-on-sapphire micro-electrode array, IEEE Trans. Electron. Devices, 26, 1932, 1979.

37. Prohaska, O. et al., A 16-fold semi-microelectrode for intracortical recordingof field potentials, Electroencephalogr. Clin. Neurophysiol., 47, 629, 1979.

38. Kuperstein, M. and Whittington, D.A., A practical 24 channel microelectrodefor neural recording in vivo, IEEE Trans. Biomed. Eng., 28, 288, 1981.

39. Shamma-Donoghue, S.A. et al., Thin-film multielectrode arrays for a cochlearprosthesis, IEEE Trans. Electron. Devices, 29, 136, 1982.

40. White, R.L. et al., Thin film electrodes for an artificial ear, J. Vac. Sci. Technol.,1, 287, 1983.

41. Takahashi, K. and Matsuo, T., Integration of multi-microelectrode and inter-face circuits by silicon planar and three-dimensional fabrication technology,Sensors Actuators, 5, 89, 1984.

42. Barth, P.W., Bernard, S.L., and Angell, J.B., Flexible circuit and sensor arraysfabricated by monolithic silicon technology, IEEE Trans. Electron. Devices, 32,1202, 1985.

43. Harrison, R.V. et al., Technical development of an implantable cochlear pros-thesis in Canada, J. Otolaryngol., 16, 311, 1987.

44. van der Puije, P.D., Pon, C.R., and Robillard, H., Cylindrical cochlear electrodearray for use in humans, Ann. Otol. Rhinol. Laryngol., 98, 466, 1989.

45. Pickard, R.S., Wall, P., and Ubeid, M., Recording neural activity in the hon-eybee brain with micromachined silicon sensors, Sensors and Actuators, B1,460, 1990.

46. Blum, N.A. et al., Multisite microprobes for neural recordings, IEEE Trans.Biomed. Eng., 38, 68, 1991.

47. Kovacs, G.T.A., Storment, C.W., and Rosen, J.M., Regeneration microelectrodearray for peripheral nerve recording and stimulation, IEEE Trans. Biomed. Eng.,39, 893, 1992.

48. Kovacs, G.T.A. et al., Silicon-substrate microelectrode arrays for parallel re-cording of neural activity in peripheral and cranial nerves, IEEE Trans. Biomed.Eng., 41, 567, 1994.

49. Kewley, D.T. et al., Plasma-etched neural probes, Sensors Actuators A, 58, 27,1997.

50. Burmeister, J.J., Moxon, K., and Gerhardt, G.A., Ceramic-based multisite mi-croelectrodes for electrochemical recordings, Anal. Chem., 72, 187, 2000.

51. Yoon, T.H. et al., A micromachined silicon depth probe for multichannelneural recording, IEEE Trans. Biomed. Eng., 47, 1082, 2000.



52. Campbell, P.K. et al., A silicon-based, three-dimensional neural interface:manufacturing processes for an intracortical electrode array, IEEE Trans.Biomed. Eng., 38, 758, 1991.

53. Jones, K.E., Campbell, P.K., and Normann, R.A., A glass/silicon compositeintracortical electrode array, Ann. Biomed. Eng., 20, 423, 1992.

54. Nordhausen, C.T., Maynard, E.M., and Normann, R.A., Single unit recordingcapabilities of a 100 microelectrode array, Brain Res., 726, 129, 1996.

55. Rousche, P.J. and Normann, R.A., A method for pneumatically inserting anarray of penetrating electrodes into cortical tissue, Ann. Biomed. Eng., 20, 413,1992.

56. Maynard, E.M., Nordhausen, C.T., and Normann, R.A., The Utah intracorticalelectrode array: a recording structure for potential brain-computer interfaces,Electroencephalogr. Clin. Neurophysiol., 102, 228, 1997.

57. Rousche, P.J. and Normann, R.A., Chronic recording capability of the Utahintracortical electrode array in cat sensory cortex, J. Neurosci. Meth., 82, 1, 1998.

58. Branner, A. and Normann, R.A., A multielectrode array for intrafascicularrecording and stimulation in sciatic nerve of cats, Brain Res. Bull., 51, 293, 2000.

59. Branner, A., Stein, R.B., and Normann, R.A., Selective stimulation of cat sciaticnerve using an array of varying-length microelectrodes, J. Neurophys., 85,1585, 2001.

60. Cheung, K. et al., A new neural probe using SOI wafers with topologicalinterlocking mechanisms, Proc. 1st Int. IEEE-EMBS Conf. on Microtechnologiesin Medicine and Biology, Lyon, 2000, p. 507.

61. Norlin, P. et al., A 32-site neural recording probe fabricated by DRIE of SOIsubstrates, J. Micromech. Microeng., 12, 414, 2002.

62. Ensell, G. et al., Silicon-based microelectrodes for neurophysiology, microma-chined from silicon-on-insulator wafers, Med. Biol. Eng. Comput., 38, 175, 2000.

63. Stieglitz, T., Beutel, H., and Meyer, J.-U., A flexible, light-weight multichannelsieve electrode with integrated cables for interfacing regenerating peripheralnerves, Sensors Actuators A, 60, 240, 1997.

64. Rodriguez, F.J. et al., Polyimide cuff electrodes for peripheral nerve stimula-tion, J. Neurosci. Meth., 98, 105, 2000.

65. Rousche, P.J. et al., Flexible polyimide-based intracortical electrode arrayswith bioactive capability, IEEE Trans. Biomed. Eng., 48, 361, 2001.

66. Meyer, J.-U. et al., High density interconnects and flexible hybrid assembliesfor active biomedical implants, IEEE Trans. Advanced Packaging, 24, 366, 2001.

67. Najafi, K. and Wise, K.D., A high-yield IC-compatible multichannel recordingarray, IEEE Trans. Electron. Devices, 32, 1206, 1985.

68. BeMent, S.L. et al., Solid-state electrodes for multichannel multiplexed intra-cortical neuronal recording, IEEE Trans. Biomed. Eng., 33, 230, 1986.

69. Bai, Q. and Wise, K.D., Single-unit neural recording with active microelec-trode arrays, IEEE Trans. Biomed. Eng., 48, 911, 2001.

70. Robblee, L.S., Lefko, J.L., and Brummer, S.B., Activated Ir: an electrode suit-able for reversible charge injection in saline solution, J. Electrochem. Soc., 130,731, 1983.

71. Anderson, D.J. et al., Batch-fabricated thin-film electrodes for stimulation ofthe central auditory system, IEEE Trans. Biomed. Eng., 36, 673, 1989.

72. Weiland, J.D. and Anderson, D.J., Chronic neural stimulation with thin-film,iridium oxide electrodes, 47, 911, 2000.



73. Hetke, J.F. et al., Silicon ribbon cables for chronically implantable microelec-trode arrays, IEEE Trans. Biomed. Eng., 41, 314, 1994.

74. Akin, T. et al., A micromachined silicon sieve electrode for nerve regenerationapplications, IEEE Trans. Biomed. Eng., 41, 305, 1994.

75. Kovacs, G.T.A. et al., Silicon-substrate microelectrode arrays for parallel re-cording of neural activity in peripheral and cranial nerves, IEEE Trans. Biomed.Eng., 41, 567, 1994.

76. Mensinger, A.F. et al., Chronic recording of regenerating VIIIth nerve axonswith a sieve electrode, J. Neurophysiol., 83, 611, 2000.

77. Najafi, K. and Hetke, J.F., Strength characterization of silicon microprobes inneurophysiological tissues, IEEE Trans. Biomed. Eng., 37, 474, 1990.

78. Edell, D.J. et al., Factors influencing the biocompatibility of insertable siliconmicroshafts in cerebral cortex, IEEE Trans. Biomed. Eng., 39, 635, 1992.

79. Najafi, K., Ji, J., and Wise, K.D., Scaling limitations of silicon multichannelrecording probes, IEEE Trans. Biomed. Eng., 37, 1, 1990.

80. Chen, J. et al., A multichannel neural probe for selective chemical delivery atthe cellular level, IEEE Trans. Biomed. Eng., 44, 760, 1997.

81. Hoogerwerf, A.C. and Wise, K.D., A three-dimensional microelectrode arrayfor chronic neural recording, IEEE Trans. Biomed. Eng., 41, 1136, 1994.

82. Bai, Q., Wise, K.D., and Anderson, D.J., A high-yield microassembly structurefor three-dimensional microelectrode arrays, IEEE Trans. Biomed. Eng., 47, 281,2000.

83. Turner, J.N. et al., Cerebral astrocyte response to micromachined silicon im-plants, Exp. Neurol., 156, 33, 1999.

84. Sinnamon, H.M. and Woodward, D.J., Microdrive and method for single unitrecording in the active rat, Physiol. Behav., 19, 451, 1977.

85. deCharms, R.C., Blake, D.T., and Merzenich, M.M., A multielectrode implantdevice for the cerebral cortex, J. Neurosci. Meth., 93, 27, 1999.

86. Venkatachalam, S., Fee, M.S., and Kleinfeld, D., Ultra-miniature headstagewith 6-channel drive and vacuum-assisted micro-wire implantation for chron-ic recording from the neocortex, J. Neurosci. Meth., 90, 37, 1999.

87. Szabo, I. et al., The application of printed circuit board technology for fabri-cation of multi-channel micro-drives, J. Neurosci. Meth., 105, 105, 2001.

88. Schmidt, E.M., McIntosh, J.S., and Bak, M.J., Long-term implants of Parylene-C coated microelectrodes, Med. Biol. Eng., 26, 96, 1988.

89. Bragin, A. et al., Gamma (40–100 Hz) oscillation in the hippocampus of thebehaving rat, J. Neurosci., 15, 47, 1995.

90. Bragin, A. et al., Multiple site silicon-based probes for chronic recordings infreely moving rats: implantation, recording and histological verification,J. Neurosci. Meth., 98, 77, 2000.

91. Swadlow, H.A. and Gusev, A.G., The influence of single VB thalamocorticalimpulses on barrel columns of rabbit somatosensory cortex, J. Neurophysiol.,83, 2802, 2000.

92. Bierer, J.A. and Middlebrooks, J.C., Auditory cortical images of cochlear-implant stimuli: dependence on electrode configuration, J. Neurophysiol., 87,478, 2002.

93. Bragin, A. et al., Gamma (40–100 Hz) oscillation in the hippocampus of thebehaving rat, J. Neurosci., 15, 47, 1995.



94. Kandel, A. and Buzsaki, G., Cellular-synaptic generation of sleep spindles,spike-and-wave discharges, and evoked thalamocortical responses in the neo-cortex of the rat, J. Neurosci., 17, 6783, 1997.

95. Buzsaki, G. and Kandel, A., Somadendritic backpropagation of action poten-tials in cortical pyramidal cells of the awake rat, J. Neurophysiol., 79, 1587, 1998.

96. Kamondi et al., Theta oscillations in somata and dendrites of hippocampalpyramidal cells in vivo: activity-dependent phase-precession of action poten-tials, Hippocampus, 8, 244, 1998.

97. Nicholson, C. and Freeman, J.A., Theory of current source-density analysisand determination of conductivity tensor for anuran cerebellum, J. Neurophys-iol., 38, 356, 1975.

98. Canning, K.J. et al., Physiology of the entorhinal and perirhinal projectionsto the hippocampus studied by current source density analysis, Ann. N.Y.Acad. Sci., 911, 55, 2000.

99. Castro-Alamancos, M.A., Origin of synchronized oscillations induced by neo-cortical disinhibition in vivo, J. Neurosci., 20, 9195, 2000.

100. Prechtl, J.C., Bullock, T.H., and Kleinfeld, D., Direct evidence for local oscil-latory current sources and intracortical phase gradients in turtle visual cortex,Proc. Natl. Acad. Sci. USA, 97, 877, 2000.

101. Ahrens, K.F. and Freeman, W.J., Response dynamics of entorhinal cortex inawake, anesthetized, bulbotomized rats, Brain Res., 911, 193, 2001.

102. Biella, G., Uva, L., and de Curtis, M., Network activity evoked by neocorticalstimulation in area 36 of the guinea pig perirhinal cortex, J. Neurophysiol., 86,164, 2001.

103. Huber, M. et al., Modification of glassy carbon surfaces with synthetic lami-nin-derived peptides for nerve cell attachment and neurite growth, J. Biomed.Mater. Res., 41, 278, 1998.

104. Cui, X. et al., Electrochemical deposition and characterization of conductingpolymer polypyrrole/PSS on multichannel neural probes, Sensors ActuatorsA, 93, 8, 2001.

105. Cui, X. et al., Surface modification of neural recording electrodes with con-ducting polymer/biomolecule blends, J. Biomed. Mater. Res., 56, 261, 2001.

106. Akin, T., Najafi, K., and Bradley, R.M., A wireless implantable multichanneldigital neural recording system for a micromachined sieve electrode, IEEEJ. Solid-State Circuits, 33, 109, 1998.

107. Takeuchi, S. and Shimoyama, I., An RF-telemetry system with shape memoryalloy microelectrodes for neural recording of freely moving insects, Proc. 1stInt. IEEE-EMBS Conf. on Microtechnologies in Medicine and Biology, Lyon, 2000,p. 491.

108. Rangarajan, R., Von Arx, J.A., and Najafi, K., Fully integrated neural stimu-lation system (FINESS), Proc. IEEE Midwest Symp. on Circuits and Systems, 3,1082, 2000.

109. Lieu, W. et al., A neuro-stimulus chip with telemetry unit for retinal prostheticdevice, IEEE J. Solid-State Circuits, 35, 1487, 2000.

110. Suaning, G.J. and Lovell, N.H., CMOS neurostimulation ASIC with 100 chan-nels, scaleable output, and bidirectional radio-frequency telemetry, IEEETrans. Biomed. Eng., 48, 248, 2001.




section four

Processing neural signals


chapter eight

Wavelet methods in biomedical signal processing

Kevin Englehart, Philip Parker, and Bernard Hudgins

Contents

8.1 Introduction8.2 A brief history of wavelets8.3 The theory of wavelets

8.3.1 The continuous wavelet transform8.3.2 The discrete wavelet transform

8.3.2.1 Subspace analogy8.3.2.2 Multiscale filter banks

8.3.3 Interpretation of the wavelet transform8.3.4 The wavelet packet transform8.3.5 Time-frequency interpretation

8.4 Wavelets in biomedical signal processing8.5 Myoelectric control applications

8.5.1 Background8.5.2 Wavelet based signal representation

8.6 Conclusions8.7 Further readingReferences

8.1 IntroductionIt has long been recognized that important features of biomedical signalsexist in both the time and frequency domains. More recently, a greater under-standing of physiological systems has been achieved by articulating therelationship between the time and frequency characteristics of biological



signals. The most widely used tool for analyzing signals in the frequencydomain is the Fourier transform.1 Fourier methods assume that the signalsof interest consist of a (perhaps infinite) summation of sinusoids. This leavesFourier analysis poorly suited to model signals with transient components.Alas, this is almost invariably the nature of signals of biological origin, inwhich the information of interest is often well localized temporally (or spa-tially). This is certainly the case in the electroencephalograph, the electrocar-diogram, the myoelectric signal, and evoked potentials of many forms.

Despite its disadvantages, Fourier methods have been widely used inthe analysis of biomedical signals. In an attempt to localize informationtemporally, the approach taken is to divide long-term signals into shorterwindows and perform a Fourier transform on each window. This is the short-time Fourier transform (STFT), which yields a time–frequency representationby expressing the frequency spectrum of each time window. By shorteningthe windows of the STFT to achieve better time resolution, however, onedegrades the resolution of the signal in frequency. Longer data windowsimprove the spectral resolution but, in degrading the temporal resolution,may violate the assumption of stationarity within each window.

In the past decade, researchers in applied mathematics and signal pro-cessing have developed wavelet methods, a powerful new framework foranalyzing transient phenomena in signals.2,15 The fundamental differencebetween wavelet and Fourier methods is the manner in which they localizeinformation in the time-frequency plane. Whereas the STFT localizes infor-mation in time–frequency cells that are of fixed aspect ratio, wavelet methodsare capable of trading off time and frequency resolution. Specifically, at lowfrequencies the representation has good frequency resolution, and at highfrequencies good temporal resolution. This tradeoff has been shown to beparticularly adept at localizing information in physical signals, particularthose of biological origin.

This chapter first provides a brief account of the development of waveletmethods, followed by an introduction to the mathematics of wavelet analy-sis. In the context of the strengths that wavelet methods have to offer, theapplication of these techniques in biomedical signal processing is described.The final section of this chapter elaborates on a wavelet-based feature setthat has resulted in robust, pattern-recognition-based control of poweredartificial limbs.

8.2 A brief history of waveletsThe concept of analyzing signals at different scales or resolutions is not new.3Only recently, however, has wavelet theory been unified, mostly due to thework of Grossman (a theoretical physicist) and Morlet (a geophysical engi-neer) in the late 1970s.4 Wavelet basis functions were shown to be effectivein localizing short, high-frequency sound waves, as well as subtle frequencychanges in long, low-frequency tones. In 1985, Meyer (a mathematician atEcole Polytechnique, near Paris) first demonstrated the construction of an



orthonormal wavelet basis with excellent time–frequency localization prop-erties.5 Shortly thereafter, Battle and LeMarie (colleagues of Meyer) came upwith an orthonormal spline wavelet using completely different techniques.6

In 1986, Mallat (a graduate student at Penn State University, a specialistin computer vision) conceived of a pyramid algorithm for wavelet decom-position and, in collaboration with Meyer, developed the theory of multires-olution analysis.7 This framework explained all the “miracles” in the waveletbases constructed up to then and made it very easy to construct neworthonormal wavelet bases. More importantly, multiresolution analysis ledto a simple and recursive filtering algorithm to compute the wavelet decom-position of a signal.

A filtering analogy of wavelet decomposition led to the formulation ofthe subband filtering approach. Here, lowpass and highpass quadraturemirror filters (QMFs) simplified the implementation of wavelet analysis andsynthesis. By considering wavelet decomposition as a series of highpass andlowpass filtering operations, wavelet packet bases were conceived8 that havegeneralized and extended the capabilities of wavelet bases.

8.3 The theory of waveletsThis section provides a brief introduction to wavelet theory. This is by nomeans meant to be an exhaustive treatment of wavelet mathematics butsimply is intended to provide a basic reference and to examine the propertiesof the wavelet transform in relation to biomedical applications. Several excel-lent texts offer a complete background for the interested reader.5,9,10,15 Fromthe point of view of the practitioner, there are fundamentally two types ofwavelet transforms:

1. Redundant transforms. These transforms include the continuous wave-let transform and wavelet frames and provide an overcomplete de-scription of the time–frequency plane. They are usually preferablefor signal analysis, feature extraction, and detection tasks, as they areshift invariant.*

2. Non-redundant transforms. These are orthogonal and biorthogonalwavelet bases. These transforms are desirable when some type ofdata reduction is preferred, or when orthogonality of the represen-tation is an important factor.

The computational complexity of a decomposition in terms of waveletbases using Mallat’s fast algorithm7 is typically orders of magnitude lessthan even the most efficient algorithms for redundant transforms.11,12 Becauseof the computational cost of redundant transforms, many researchers have

* We may say that a transform is translation invariant if, for a simple time shift in the inputsignal, the transform coefficients experience the same simple shift. Due to the aliasing thatoccurs in a discrete wavelet decomposition, the transform is not translation invariant.



applied non-redundant transforms to signal analysis, feature extraction, anddetection tasks and have obtained satisfactory results.

8.3.1 The continuous wavelet transform

The continuous wavelet transform (CWT) provides a variable coverage of thetime

-frequency plane. The transform is defined as:

where

where

ψ(t) is a prototype window referred to as the mother wavelet. Themother wavelet has the property that the set

ψa,

τ(t)a,

τ∈Z forms an orthonor-mal basis in L2(

ℜ), the space of square-integrable functions.* The analysisdetermines the correlation of the signal with shifted (by

τ) and scaled (by a)versions of the mother wavelet.

The parameter scale in the wavelet analysis is similar to the scale used inmaps. As in the case of maps, high scales correspond to a non-detailed globalview (of the signal), and low scales correspond to a detailed view. Similarly,in terms of frequency, low frequencies (high scales) correspond to a globalinformation of a signal (that usually spans the entire signal), whereas highfrequencies (low scales) correspond to a detailed information of a hiddenpattern in the signal (that usually lasts a relatively short time). Figure 8.1demonstrates the nature of a symmetric wavelet at various scales and trans-lations. Note that, at small scales, a temporally localized analysis is done; asscale increases, the breadth of the wavelet function increases, thereby analyz-ing with less time resolution but greater frequency resolution. Notice, as well,that the wavelet functions are bandpass in nature, thus partitioning the fre-quency axis. In fact, a fundamental property of wavelet functions is that

where

∆f is a measure of the bandwidth, f is the center frequency of the passband,and c is a constant. The wavelet functions may therefore be viewed as a bankof analysis filters with a constant relative passband (a

“constant-Q

” analysis).

* A function f is in L2(

ℜ) if .

CWT x aa x t t dt( , ) ( ) ,*τ ψ τ= ( )∫

ψ ψ ττa t

at

a, ( ) =1 −

L f2 2( ) ℜ < ∞ℜ∫if

cf

f =

∆



8.3.2 The discrete wavelet transform

Fast wavelet transforms are obtained through multiresolution analysis; a pyra-mid algorithm with its origins in image processing that was adapted forwavelet analysis by Mallat and Meyer.7 The fast wavelet transform uses aseries of linear filters — lowpass and highpass — to decompose the signalinto low- and high-frequency components. The algorithm also combines thesefilters with downsampling operations, that is, steps that decimate the signal ateach stage, halving the data each time. This feature accounts for the algorithm

’sspeed, because the downsampling reduces the computations at each iterationgeometrically — at j iterations, the number of samples being manipulatedshrinks by 2j. This yields very efficient algorithms: most N-point wavelettransforms have complexity on the order of O(N), whereas a Fourier transformis of the order O(NlogN).13 If the transform

’s complexity is CN (where C is aconstant), then C depends on the wavelet chosen.14 If C is small, then comput-ing the wavelet transform requires about the same effort as trivial tasks suchas copying or rescaling a signal. Wavelets involving only a few terms subtendgreat efficiency (a small value of C), such as those developed by Daubechies.15

In its discrete form, a = a0j and

τ = n

⋅ a0–j where j and n are integers; this

is referred to as the discrete wavelet transform (DWT). In this context, j controlsthe dilation or compression of the wavelet function, and n controls the trans-lation in time. If we choose a0

≈ 1 and n

≈ 0, we are close to the continuouscase. The choice of a = 2j and

τ = n

⋅ 2–j is the most common choice; this isreferred to as a dyadic wavelet basis. The dyadic form also simplifies the con-straints on the mother wavelet to achieve another desirable property: orthog-onality of the analysis windows, which subtends an efficient representation.

Figure 8.1 Some symmetric wavelets at various scales and locations. The couplet (a,

τ) marking each waveform denotes scale and shift, respectively: (a) time domain and(b) frequency domain of each wavelet.



One can interpret the discrete wavelet transform in a number of ways,but the most common are via subspace analogy and multiscale filter banks.

8.3.2.1 Subspace analogyIn the dyadic case, the mother wavelet is

and the set

ψj,n(t)j,n

∈Z forms a sparse orthonormal basis of L2(

ℜ).16 Thismeans that the wavelet basis induces an orthogonal decomposition of anyfunction in L2(

ℜ):

where

Ωj,1 is the subspace spanned by

ψj,n(t)n∈Z and ⊕ is the direct sum ofthe subspaces. Thus, a complete description of the original signal is availablefrom a direct sum of orthogonal subspaces. The subscript 1 is used to dif-ferentiate Ωj,1 from its dual space Ωj,0, which is defined as follows. A waveletfunction is always associated with a companion: the scaling function, ϕ(t),which is also sometimes called the father wavelet:

and, like the wavelet function, the set ϕj,n(t)n∈Z forms a sparse orthonormalbasis of L2(ℜ). The scaling function induces a chain of nested subspaces:

where Ωj,0 is the subspace spanned by ϕj,nn∈Z.What does this really mean? The nature of the scaling function is that a

projection of the original signal x(t) onto the space Ωj,0 is a lowpass operation.Specifically, the projection of x(t) onto Ωj,0 is an approximation at scale a = 2j:

where aj,n = ⟨x(t), ϕj,n(t)⟩ are the scaling coefficients. We define Ω0,0 (scale a = 20

= 1) to be the space of the original signal x(t); that is, A0[n] = x[n]. Thus, ΩJ,0

⊂ ΩJ–1,0 ⊂ … ⊂ Ω1,0 ⊂ Ω0,0 is actually a sequence of successively coarserapproximations of x(t) as scale ranges from 0 to J. The subspaces subtendedby the wavelet and the scaling functions are related such that:

ψ ψj nj jt t n, ( ) ( )= −− −2 22

Lj

j2

1( ) ,ℜ = ⊕Ω

ϕ ϕj nj jt t n, ( ) ( ) 2= −− −2 2

Ω Ω Ω ΩJ J, , , ,0 1 0 1 0 0 0⊂ ⊂ ⊂ ⊂− L

A n a tj j nn Z

j n[ ] ( ), , =∈∑ ϕ



meaning that Ωj,1 contains the detail needed to go from a coarser to a finerlevel of approximation. The detail component of x(t) at scale a = 2j is:

where dj,n = ⟨x(t), ψj,n(t)⟩ are the wavelet coefficients. This is a bandpass operation,as DJ[n] = AJ+1[n] – AJ[n]. Therefore, the wavelet transform may be viewed asa way to represent Ω0,0 as a direct sum of mutually orthogonal subspaces:

This concept is perhaps more clearly illustrated in Figure 8.2. The wavelettransform processes a signal by decomposing it into successive approxima-tion Aj[n] ∈ Ωj,0 and detail Dj[n] ∈ Ωj,1 signals. The approximation signal isresampled at each stage, and the detail coefficients are kept. The aim of theanalysis is to arrive, starting from the original sampled signal x[n] = A0[n],at a decomposition into detail and approximation signals:

This is the wavelet transform: for a decomposition into J scales, the transformcoefficients consist of J scales of detail coefficients and, at the Jth scale, thelowest-level approximation signal.* Equivalently, one may retain the waveletand scaling coefficients which generate these signals:

* This is illustrated by example in Figure 8.5, in the section “Interpretation of the Wavelet Transform.”

Figure 8.2 Subband decomposition analogy of the wavelet transform. The symbolssurrounded by gray represent those subspaces kept intact by the wavelet transform.

Ω Ω Ωj j j j J, , , , , ,0 1 0 1 1 0 1 for= ⊕ =+ + K

D n d tj j nn Z

j n[ ] ( ), , =∈∑ ψ

Ω Ω Ω0 01

1 0, , , =

=

⊕⊕

j

J

j J

D n D n A nJ J1[ ], , [ ], [ ]K

Ω

Ω

Ω

Ω

Ω

Ω

Ω 0 0,

1 1,

1,0

2 0,

2 1,

J ,1

J ,0



8.3.2.2 Multiscale filter banksFor those so inclined, it is instructive to repeat the theory of wavelet decom-position in the language of signal processing. Let the map Ωj,0 → Ωj+1,0 berepresented by the operator H, and the map Ωj,0 → Ωj+1,1 be represented byG. This is illustrated in Figure 8.3 as an operation on an N-sample signal x= .

As the figure implies, the operators consist of two separate stages: and are highpass and lowpass filters of length

L, respectively, and the ↓2 operator implies a decimation by 2. This may bewritten as:

for n = 0, 1, …, N – 1. This makes intuitive sense, that the approximationΩj,0 → Ωj+1,0 should be obtained via lowpass filtering, and the detail Ωj,0

→ Ωj+1,1 by highpass filtering. If is a vector to be ana-lyzed, the operators transform the vector x into two subsequences Gxand Hx, of length N/2. Next, the same operations are applied to the vectorof the lower frequency band Hx to obtain H2x and GHx of lengths N/4.If the process is repeated J ≤ log2N times, the wavelet decomposition maybe then written as:

of length N. This is demonstrated in Figure 8.4. The wavelet transformthus analyzes the data by partitioning its frequency content dyadicallyfiner and finer toward the low frequency region (and coarser and coarserin the time domain).

The issue to be addressed now is, how are the filters h and g determined?The operators H and G are called perfect reconstruction or quadrature mirrorfilters (QMFs) if they satisfy the following orthogonality conditions:

Figure 8.3 Projection operators H and G.

d d an J n J n1, , ,, , ,K

x n n

N[ ] , ∈=−01

0 0Ω

g = [ ] =0

-g n

n

L 1h = [ ] =

-h nn

L

0

1

H h k x n k G g k x n knk

L

nk

L

x x( ) = − ( ) = −=

−

=

−

∑ ∑ 0

1

0

1

2 2[ ] [ ], [ ] [ ],

x = ∈ℜ=−x n n

N N[ ] 01

, , , , , G GH GH GH HJ Jx x x x x2 1K +



where * is the adjoint operator and I is the identity matrix. Various designcriteria (concerning regularity, symmetry, etc.) on the lowpass filter coeffi-cients h can be found in Daubechies.15 Once the h are fixed, we can specifythe QMFs by setting g[n] = (–1)nh[L – 1 – n].

The QMFs are related to the wavelet and scale functions by:16

and

If H and G are QMFs, then the perfect reconstruction property allowsexact reconstruction of the original signal from the wavelet transformcoefficients.17 Because the wavelet bases are orthogonal, reconstructiontakes the form of upsampling (inserting zeros at every other sample) andfiltering by the QMFs, exactly the reverse operation of the subbanddecomposition scheme.*

8.3.3 Interpretation of the wavelet transform

After the mathematic details presented above, an example might clarifythings a bit. Consider a N = 1024 point ramp function, depicted in Figure 8.5a.The ramp is then subjected to a multiresolution analysis using a Symmlet-5wavelet, decomposed to level 7 (the decomposition could have been carriedout to a maximum depth of J = log21024 = 10, but this would have unnec-

Figure 8.4 The subband coding analogy of the DWT.

* For orthogonal wavelet bases, the filters g and h are used for both decomposition and synthesis.An alternative scheme, referred to as biorthogonal wavelet bases, allows perfect reconstructionwith synthesis filters that are different than the analysis filters.

HG GH HH GG* * * *= = = =0 and I

h n t t n[ ] ( ), ( )= −ϕ ϕ2 2

g n t t n[ ] ( ), ( )= −ψ ϕ2 2



essarily crowded the plot*). Clearly, the detail signal D1[n] picks up thesharpest detail of the step transition (the highest bandpass); the detail signalsfrom D2[n] to D7[n] describe the signal content in bandpass regions ofsuccessively lower center frequency. The approximation signal A7[n] at level7 provides the lowpass estimation of the signal.

This decomposition can also be represented by the wavelet coefficientsthemselves, as shown in Figure 8.6. Note the decimation that occurs whenprogressing from d1,n to d7,n. There are N/2 = 512 coefficients in d1,n, N/4 =256 coefficients in d2,n, and so on, until at level 7, d7,n and a7,n have eightcoefficients each. This yields a total of 1024 coefficients, the same as theoriginal signal. Due to the perfect regeneration properties of the QMFs, theoriginal signal can be exactly reconstructed from the wavelet coefficients.Note that signal filtering, denoising, or compression is possible by comput-ing the wavelet coefficients and then modifying or eliminating some beforereconstructing the signal.

8.3.4 The wavelet packet transform

The wavelet packet transform5,18–20 is a generalized version of the wavelettransform; it retains not only the low- but also the high-frequency subband,performing a decomposition upon both at each stage. As a result, the tilingof the time-frequency plane is configurable: the partitioning of the frequency

Figure 8.5 (a) A N = 1024 point ramp function, and (b) its multiresolution waveletanalysis.

* Although a decomposition can carried out to its maximal depth J = log2N, the decompositionmay terminate at any level. This means that the process of dividing the frequency axis intofiner and finer segments toward the low frequency range stops, and the final detail and ap-proximation signals are of length greater than one. This may be desirable if the subdivision ofbands beyond a certain scale does not yield subbands with a significant energy component.



axis may take many forms to suit the needs of the application. This isdemonstrated in Figure 8.7.

Starting with a signal x of length N samples, the first level of thedecomposition generates the lowpass and highpass subbands (Hx andGx, respectively) as with the wavelet transform, each of half the lengthof x. The second level decomposition generates four subsequences: H2x,GHx, HGx, and G2x halved in length again; the decomposition to thislevel is shown in Figure 8.8.

Figure 8.6 The wavelet coefficients of the ramp function shown in Figure 8.5a. De-composition has been carried out to level 7.

Figure 8.7 The time-frequency plane tiling of (a) a wavelet basis, and (b) an arbitrarywavelet packet basis.

Figure 8.8 The first two levels of decomposition in a wavelet packet transform. Thelowpass operations (H) occur to the left, the highpass (G) to the right.

0 256 512 768 1024Time (samples)

Wavelet Coefficients

nd .1

nd .2

nd .3

nd .4

nd .6

nd .7

na .7

(a) (b)

TimeTime

Fre

quen

cy

Fre

quen

cy

xxH xG

x2H xGH xHG x2G



This process is repeated J times, where J

≤ log2N, resulting in JN coeffi-cients. For a subband at scale j, there are N/2j wavelet packet coefficientssampled at fs/2j spanning the entire time record. This iterative process gen-erates a binary wavelet packet tree structure where the nodes of the tree rep-resent subspaces with different frequency localization characteristics.

As an example, consider a sinusoid with a frequency that increases linearlywith time (a linear chirp). The binary wavelet packet tree of this signal isdepicted in Figure 8.9. Each subband in the full decomposition is separatedby dashed lines; the wavelet packet coefficients within each subband areshown as solid lines. The vertical axis indicates the depth of decomposition;the zero scale is actually the original signal. The horizontal axis is labeledFrequency [Time], implying that, for a given scale (level), the center frequencylocalization of each subband increases as one progresses from left to right.

The time-frequency localization of each subband is evident; within eachsubband, the significant coefficients are temporally localized in the rangewhere the chirp passes through the frequency range corresponding to thatsubband. The computational cost of this decomposition is on the order ofO(JN)

≤ O(Nlog2N).20

In subspace notation, the root node of the tree is

Ω0,0. The node

Ωj,k isdecomposed into two orthogonal subspaces H:

Ωj,k

→

Ωj+1,2k and G:

Ωj,k

→

Figure 8.9 The binary wavelet packet tree for a 256-sample chirp signal. A Coiflet-5 wave-let was used, which provides a good tradeoff between time and frequency localization.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

5

4

3

2

1

0

Frequency [Time]

Sca

le



Ωj+1,2k+1. Here, j denotes scale, as before, and k indicates the subband indexwithin the scale.* This may be expressed as:

for j = 0, …, J and k = 0, …, 2j – 1, which generates the entire decompositionto scale J. A decomposition to scale J = 3 is shown in Figure 8.10.

The set of subspaces in the binary tree is a redundant set. Indeed, thereare more than possible orthonormal bases in this binary tree.20 Eachsubspace Ωj,k is spanned by basis vectors . In a givenfamily of wavelet packet bases, the parameter j indicates scale, as before.The parameters k and n roughly indicate frequency band** and the centerof the waveform, respectively. The vector wj,k,n is roughly centered at 2jn, haslength of support ≈2j, and oscillates ≈k times. For j = 0, we have the originalsignal space (the standard Euclidean basis [ℜN]).

For a J-scale decomposition, the resulting binary tree yields more than orthonormal bases (or coordinate systems), all of which offer a com-

plete description of the space of the original signal. The power of the waveletpacket transform is that a “best basis” can be chosen for a specific task, if itcan be properly identified from the ensemble of possible candidates.

To determine the best basis, it is necessary to evaluate and compare theefficacy of many bases. To this end, a cost function must be chosen to represent

Figure 8.10 A decomposition of Ω0,0 into binary tree-structured subspaces using thewavelet packet transform (with J = 3).

* The wavelet transform has only two subbands per scale, high and low, with k = 0, 1.** The binary tree subband structure, as generated by successive applications of H and G, iscalled Paley or natural ordering. In this form, the frequency band of Ωj,k does not monotonicallyincrease with k. This may be corrected by Gray-code permutation.22

Ω Ω Ωj k j k j k, , ,= ⊕+ + +1 2 1 2 1

22 1( )J−

2 0n j− wj k n n

n j

, , =−−

0

2 10

22 1( )J−



the goal of the application. The best-basis selection algorithm has its originsin signal compression,20,21 and the cost functions associated with compressionall entail some form of entropy measure. Saito22 has described a modificationto the best-basis algorithm to suit classification problems. Termed the localdiscriminant basis algorithm, it replaces the entropy measure with distancemeasures between classes of signals.

8.3.5 Time-frequency interpretation

One of the most important ways of interpreting the information content ina signal is through its time–frequency response (TFR). The time–frequencyrepresentation tends to be more intuitive than time scale, perhaps becauseof the historical use of time–frequency methods such as the short-time Fou-rier transform and the Wigner–Ville distribution. The TFR of the wavelettransform is computed by translating scale into frequency. The generationof the TFR of a wavelet transform is a frequent requirement of many inves-tigators, and it is surprising that the mechanics of this task are overlookedin many texts.

Consider a signal of length (n0 = 3) sampled at fs Hz. Thetiling of the time-scale grid (with a wavelet decomposition to scale J = log2N= 3) will look like that shown in Figure 8.11a. The selection of a dyadicsampling grid (a = 2j, τ = n2j) means that the resulting TFR will be criticallysampled; there will be N discrete time–frequency cells.

For a given information cell, the temporal resolution is ∆t = 2j ⋅ Ts, andthe scale resolution is ∆a = 2J–j. The subscript 1 or 0 on the scale parameterdenotes whether the subband is detail (bandpass) or approximation (lowpass).As expected, the lowest frequency subband is composed of the approxima-tion coefficients from the highest scale level. At scale level j, there are((NTs)/(∆t)) = N ⋅ 2–j = information cells. The subband at each scale j< J is described by detail coefficients. At scale J, one bandpass subbandis described by detail coefficients, and one lowpass subband isdescribed by approximation coefficients.

Figure 8.11 The tiling of the time-scale domain (a) and the time-frequency domain(b) of the DWT. In (a), the scale is subscripted by a 1 for detail subbands and 0 forthe approximation subband.

N n= =2 80

2 0n j−

2 0n j−

2 0n J−

2 0n J−



Translating this time-scale response into a time–frequency response pro-ceeds as follows. Referring now to Figure 8.11b, the temporal resolution ofa given cell is still ∆t = 2j ⋅ Ts. To compute the frequency segmentationcorresponding to scale, we may identify the vertical distance from 0 Hz tothe bottom of each cell as f, and the height of that cell as ∆f. On this grid, ∆f= (∆a/2J)(fs/2) = 2–j(fs/2). The distance f to the lower bound of a cell is givenby f = ∆f ⋅ q, where q = 1 if the subband consists of detail coefficients and q= 0 if it is an approximation subband. Here, we have chosen a definition of∆t and ∆f so as to yield a critically sampled time–frequency grid. It is inter-esting to note that ∆t ⋅ ∆f = (2J ⋅ Ts)(2–jfs/2) = 1/2, which is somewhat greaterthan the minimum set by the Heisenberg bound. A chosen wavelet therefore,need not meet the Heisenberg bound but rather must satisfy ∆t ⋅ ∆f ≤ 1/2 toavoid overlap amongst adjacent cells in the TFR.* This critically sampledgrid also ensures perfect reconstruction of the original signal. The details ofconstructing the TFR of the wavelet packet transform are somewhat moreinvolved than the wavelet transform. For details, the reader is referred toSimoncelli et al.23

8.4 Wavelets in biomedical signal processingThe reader should be cautioned that, as with many tools, there have beeninappropriate uses of the wavelet transform:

“When all you have is a hammer, everything is a nail.”

For the most part, however, there have been some remarkable advances inbiomedical signal processing that exploit the capabilities of the wavelet andwavelet packet transform. An important aspect of biomedical signals is thatthe information of interest is often a combination of phenomena that aretransient (e.g., spikes and action potentials) and diffuse (e.g., small oscilla-tions, and semiperiodic activations). This requires analysis methods that aresufficiently versatile to capture events that present these extremes intime–frequency localization. It is this ability of the wavelet transform thathas motivated its use in biomedical applications.

The popularity of wavelets in biomedicine is evidenced by the steadilygrowing number of publications: two special issues published in IEEE Engi-neering in Medicine and Biology Magazine (March/April 1995 and October1995), workshops at many biomedical engineering conferences, and someexcellent texts24,25 and review papers26,27 specifically addressing biomedicalapplications of wavelets.

The continuous wavelet transform (CWT), in essence, performs a corre-lation analysis upon the signal of interest. Correspondingly, one can expect

* Most wavelets transforms satisfy ∆t ⋅ ∆f ≤ 1/2 sufficiently; any overlap is not severe enoughto distort the TFR.



its output to be a maximum when the signal most resembles the motherwavelet ψa,τ(t). In this sense, the CWT behaves as a multiscale matched filter.This property of the CWT has motivated its use as a detector, looking for apattern (or elements of a pattern) that may be scaled by various amounts.The CWT has been used in this context for detecting microcalcifications inmammograms28 and the QRS complex in the ECG signal.29

The ability of the wavelet transform to offer good frequency resolutionat low frequencies and good time resolution at high frequencies has madeit appealing for the time–frequency analysis of many biological signals.Recent examples of this include the characterization of heart beat sounds,30,31

the detection of ventricular late potentials in the ECG,32–35,50 and the analysisof the EEG.36–38

The efficiency of representation offered by orthogonal wavelet trans-forms has made them an attractive choice in data compression and noiseremoval; this can be achieved by modifying or discarding certain waveletcoefficients that are insignificant.39 Indeed, wavelet packet based compres-sion has set the standard for fingerprint image compression, offering a 20:1compression ratio, as compared to the 5:1 ratio offered by JPEG encoding.14

Orthogonal wavelet decomposition has been found to be very useful inimage coding40 and compression of magnetic resonance images41 and dig-ital mammograms,56 the ECG,42,43 and the myoelectric signal.44,45 They havealso been used for noise removal in evoked response potentials46 and inthe ECG.47

Wavelet representations have also been used as feature sets in patternrecognition problems. The problem of characterizing ECG patterns is ofenormous interest; it is not surprising that attempts have been made to usewavelet-based feature sets. It has been shown that wavelet features out-perform a set of morphological features extracted from a quadratictime–frequency distribution when classifying heart sounds.48 Others havesuccessfully used Mallat’s local extrema representation49 to recognize car-diac patterns.50,51 Wavelet coefficients have been used for identification ofrespiratory-related evoked potentials,52 image texture segmentation,53 andclassification of magnetic resonance images.54 Wavelet and wavelet packetcoefficients have also been successfully used in myoelectric signal classifi-cation for control of prosthetic limbs.55,56

Wavelet transforms have also been used for characterization of soma-tosensory event related potentials for tracking cerebral hypoxia,57 pitchdetection of speech signals,58 and many applications in biomedical imaging.26

This list is by no means comprehensive, but rather represents the diversityof wavelet applications in biomedicine.

The next section elaborates upon the use of the wavelet transform inpattern recognition. The control of powered artificial limbs using the myo-electric signal requires a representative feature set to be computed in realtime. The computational efficiency of the orthogonal wavelet transform, andits ability to model transient myoelectric signal patterns have resulted in acontrol system of greater accuracy than its predecessors.



8.5 Myoelectric control applications8.5.1 Background

The myoelectric signal (MES), recorded at the surface of the skin, has beenused for many diverse applications, including clinical diagnosis, and as asource of control of assistive devices and schemes of functional electricalstimulation. This section details recent work that improves the functional-ity and ease of control of powered upper-limb prostheses using the myo-electric signal.

Many myoelectric control systems are currently available that are capa-ble of controlling a single device in a prosthetic limb, such as a hand, anelbow, or a wrist. These systems extract control information from the MESbased on an estimate of the amplitude59 or the rate of change60 of the MES.Although these systems have been very successful, they do not providesufficient information to reliably control more than one function (or device);61

the extension to controlling multiple functions is a much more difficultproblem. Unfortunately, these are the requirements of those with high-level(above the elbow) limb deficiencies, and these are the individuals who couldstand to benefit most from a functional replacement of their absent limbs.

In an attempt to increase the information extracted from the MES, inves-tigators have proposed a variety of feature sets and have utilized patternrecognition methods to discriminate amongst desired classes of limb activa-tion. Most work in MES classification has considered the steady-state MES,collected during a maintained (usually constant-force) contraction. Hudginset al.62 were the first to consider the information content of the transientbursts of myoelectric activity that accompany the onset of contraction. Thesedata were acquired in a single MES channel, using a widely spaced bipolarelectrode pair placed on the biceps and triceps. The data were acquired bytriggering on an amplitude threshold of a moving average of the absolutevalue of the transient waveforms. The structure inherent in the early portionof these transient bursts (roughly the first 100 msec) suggested a promisingmeans of MES classification. Hudgins developed a control scheme basedupon a set of simple time domain statistics and a multilayer perceptronartificial neural network classifier, capable of classifying four types of upperlimb motion from the MES acquired from the biceps and triceps. This controlscheme demonstrated greater discriminant ability than any other at the timeand allowed a user to evoke control using muscular contractions that resem-ble those normally used to produce motion in an intact limb. This systemhas been implemented as an embedded controller63 and is currently under-going clinical trials.

Although the accuracy of Hudgins’ controller is good (roughly 10% error,averaged over a set of 10 subjects), there is an obvious motivation to reducethe error as much as possible. This would enhance the usability of the systemas perceived by the user and allow greater dexterity of control. A numberof approaches have appeared in the literature that have used the transient



signal as prescribed by Hudgins, seeking to improve the accuracy of theapproach using dynamic artificial neural networks,64 genetic algorithms,65

fuzzy logic classifiers,66 and self-organizing neural networks.67

8.5.2 Wavelet based signal representation

Instead of focusing upon the classifier, the authors have demonstrated inprevious work that the classification performance is more profoundlyaffected by the choice of feature set.56 Specifically, a wavelet-based approachis described that, in direct comparison to Hudgins’ time-domain approach,exhibits superior performance. The performance of Hudgins’ time-domain(TD) feature set, and those based upon the short-time Fourier transform(STFT), the wavelet transform (WT), and the wavelet packet transform (WPT)were compared using a new dataset. This work is briefly described here.

A roster of 16 healthy subjects participated in the study. Each subjectgenerated six different classes of motion: closing/opening the hand, flex-ion/extension of the wrist, and ulnar/radial deviation of the wrist. The datawere acquired from four channels located on the medial side, top, lateralside, and bottom of the forearm, as shown in Figure 8.12. Each patternconsists of two channels of 256 points, sampled at 1000 Hz. The data weredivided into a training set (100 patterns) and a test set (150 patterns).

Each of the STFT, WT, and WPT implementations were empirically opti-mized to yield the best possible classification performance from the ensembleof 16 normally limbed subjects. For the STFT, it was determined that (froma number of taper windows) a Hamming window of length 64 points withan overlap of 50% gave the best performance. When using the WT, a Coifletmother wavelet (of order four) yielded better accuracy than a host of otherwavelet families of varying order.15 The WPT experienced the best perfor-mance when using a Symmlet mother wavelet (of order five). A number ofmethods were considered as candidates to determine the best tiling of theWPT. The most common approach to specifying the WPT tiling is by selectingthat which minimizes the reconstruction error, using an entropy cost func-

Figure 8.12 The electrode placement used in the four-channel MES acquisition. Fourbipolar electrode pairs (Red Dot®; 3M Corp.) were used with a reference at the wrist.Although difficult to show on the figure, the top and bottom electrode pairs are atthe same level of the forearm.

Channel 2 Channel 1 Channel 4(back)

Reference Channel 3



tion.68 This may be considered optimal for signal compression but may beinappropriate for signal classification. A modified form of this algorithm,called the local discriminant basis algorithm, has been proposed that seeks tomaximize the discriminant ability of the WPT by using a class separabilitycost function.22 It is established in Englehart et al.56 that this discriminantcost function does indeed produce the best classification performance.

From each subject, the TD, STFT, WT, and WPT feature sets werecomputed. Also considered was the stationary wavelet transform (SWT),which does not decimate the signal at each stage, as does the standarddiscrete WT.69 Subsequently, each feature set was subject to dimensionalityreduction using principal components analysis (PCA), so as not to over-whelm the classifier with high-dimensional data. It is shown in Englehartet al.56 that the application of PCA is critical to the success of the time–fre-quency-based feature sets, and that PCA is clearly superior to other formsof dimensionality reduction. Although the classification performance is notsensitive to the dimensionality of the PCA-reduced feature set, it wasdemonstrated that at least 5 PCA features are needed, and more than 30unnecessarily burdens the classifier; 20 PCA coefficients are used in theanalyses described here.

Figure 8.13 shows the classification error for each of the candidate featuresets, averaged over all 16 subjects, using a linear discriminant analysis clas-sifier. Clearly, the time-domain feature set performs very well with fourchannels of MES (as compared to one used by Hudgins), with an error ofroughly 4.8%. The STFT does not fare as well, most likely due to its inabilityto efficiently capture the energy of the signal in the time–frequency plane.The performance of the WT approaches that of the TD feature set, and theWPT surpasses it, due to the adaptive nature of the time–frequency tilingafforded by the local discriminant basis algorithm. The stationary wavelet

Figure 8.13 The classification error associated with each feature set, averaged acrossall 16 subjects.

TD STFT WT WPT SWT3.5

4

4.5

5

5.5

6

6.5

7

Feature Set

Err

or (

%)



transform performs best of all, due to the fact that it is translation invariant.*Added computational expense is associated with the SWT, but this is notenough to preclude a real-time implementation on modern microprocessors.This improved classification accuracy directly translates into more robustcontrol of a prosthetic limb.

8.6 ConclusionsThis introductory view of wavelets provides the practitioner with a briefhistory, the basic theory, and the fundamental motivation for their use inbiomedical signal processing. It should be respected that the use of waveletsshould be motivated by their strengths and an understanding of the biologyand physics of the problem under consideration, rather than the fact thattheir use may be considered fashionable. Nonetheless, many advances attrib-utable to wavelets have been made in the past decade, as they are shown tooutperform the best techniques otherwise available.

8.7 Further readingReviews of wavelet theory as applied to signal processing in general can befound in References 1, 11, 70, and 71. Other information beyond the scopeof the discussion here includes wavelet types,16 biorthogonal transforms,38,72

translation invariant methods,73–79 and matching pursuit.80 By far, the greatestresource for wavelet information is the Internet. It is impossible to summa-rize the diversity of information available online, but the reader is directedto some “meta” sites that provide direction to literature, software, and dis-cussion groups. These include:

http://www.amara.com/current/wavelet.htmlhttp://www.wavelet.orghttp://www.mathtools.net/Applications/DSP/Wavelets/

References1. Akay, M., Detection and Estimation of Biomedical Signals, Academic Press, San

Diego, CA, 1995.2. Vetteri, M. and Kovacevic, J., Wavelets and Subband Coding, Prentice-Hall,

Englewood Cliffs, NJ, 1995.3. Haar, A., Zur theorie der orthogonalen funktionensysteme [in German], Math

Anal., 69, 331–371, 1910.4. Grossman, A. and Morlet, J., Decomposition of Hardy functions into square

integrable wavelets of a constant shape, SIAM J. Math Anal., 15, 723–736, 1984.5. Meyer, Y., Wavelets, Algorithms and Applications, SIAM, 1993.

* The manifestation of a lack of translation invariance in the WT and WPT is a nonlineardistortion of the coefficients with temporal shift, due to the critically sampled nature of thedyadic decomposition.



http://www.amara.com/current/wavelet.html

http://www.wavelet.org/

http://www.mathtools.net/Applications/DSP/Wavelets/index.html

6. Battle, G., A block spin construction of ondelettes. I. Lemarie functions, Comm.Math. Phys., 110, 601–615, 1987.

7. Mallat, S., A theory for multiresolution signal decomposition: the waveletrepresentation, IEEE Trans. Patt. Anal. Mach. Intel., 11, 674–693, 1989.

8. Coifman, R.R. and Wickerhauser, M.V., Entropy–based algorithms for bestbasis selection, IEEE Trans. Inform. Theory, 38(2), 713–719, 1992.

9. Kaiser, G., A Friendly Guide to Wavelets, Birkhäuser, Boston, MA, 1994.10. Strang, G. and Nguyen, T., Wavelets and Filter Banks, Wellesley-Cambridge

Press, Wellesley MA, 1997.11. Rioul, O. and Duhamel, P., Fast algorithms for discrete and continuous wave-

let transforms, IEEE Trans. Inform. Theory, 38, 569–586, 1992.12. Unser, M., Fast Gabor-like windowed Fourier and continuous wavelet trans-

forms, IEEE Signal Proc. Lett., 1, 76–79, 1994.13. Sweldens, W., Wavelets: what next?, Proc. IEEE, 84(4), 680–688, 1996.14. Bruce, A., Donoho, D., and Gao, H.–Y., Wavelet analysis, IEEE Spectrum,

October, 26–35, 1996.15. Daubechies, I., Ten lectures on wavelets, in CBMS–NSF Regional Conference

Series in Applied Mathematics, Vol. 61, SIAM, Philadelphia, 1992.16. Meyer, Y., Wavelets and Operators, Vol. 37, Cambridge Studies in Advanced

Mathematics, Cambridge University Press, Cambridge, MA, 1993 (translatedby D.H. Salinger).

17. Cohen, A., Daubechies, I., and Feauveau, J.C., Biorthoginal bases of compactlysupported wavelets, Communications of Pure and Applied Math, 1990.

18. Coifman, R.R. and Meyer, Y., Nouvelles bases orthonormées de L2(ℜ) ayantla structure du systèm de Walsh, Department of Mathematics, Yale University,New Haven, CT, 1989.

19. Coifman, R.R. and Meyer, Y., Orthonormal Wave Packet Bases, Departmentof Mathematics, Yale University, New Haven, CT, 1990.

20. Wickerhauser, M.V., Adapted Wavelet Analysis from Theory to Software, AK Pe-ters, Ltd., Wellesley, MA, 1994.

21. Coifman, R.R. and Wickerhauser, M.V., Entropy–based algorithms for bestbasis selection, IEEE. Trans. Inform. Theory, 38(2), 713–719, 1992.

22. Saito, N. and Coifman, R.R., Local discriminant bases and their applications,J. Math. Imaging Vision, 5(4), 337–358, 1995.

23. Simoncelli, E.P., Freeman, W.T., Adelson, E.H., and Heeger, D.J., Shiftablemultiscale transforms, IEEE Trans. Inform. Theory, 38(2), 587–607, 1992.

24. Aldroubi, A. and Unser, M., Wavelets in Medicine and Biology, CRC Press, BocaRaton, FL, 1996.

25. Akay, M., Time Frequency and Wavelets in Biomedical Signal Processing, JohnWiley & Sons, New York, 1997.

26. Unser, M. and Aldroubi, A., A review of wavelets in biomedical applications,Proc. IEEE, 84(4), 626–238, 1996.

27. Akay, M., Wavelets in biomedical engineering, Ann. Biomed. Eng., 23, 531–542,1995.

28. Strickland, R.N. and Hahn, H.I., Detection of microcalcifications in mammo-grams using wavelets, in Proc. SPIE Conf. Wavelet Applications in Signal andImage Processing II, San Diego, CA, Vol. 2303, 1994, pp. 430–441.

29. Li, C. and Zheng, C., QRS detection by wavelet transform, Proc. Annu. Conf.Eng. Med. Biol., 15, 330–331, 1993.



30. Khadra, L., Matalgah, M., El–Asir, B., and Mawagdeh, S., The wavelet trans-form and its applications to phonocardiogram signal analysis, Med. Informat-ics, 16(3), 271–277, 1991.

31. Obaidat, M.S., Phonocardiogram signal analysis: techniques and perfor-mance, J. Med. Eng. Technol., 17, 221–227, 1993.

32. Khadra, L. Dickhaus, H., and Lipp, A., Representations of ECG late potentialsin the time–frequency plane, J. Med. Eng. Technol., 17(6), 228–231, 1993.

33. Dickhaus, H., Khadra, L., and Brachmann, J., Time–frequency analysis ofventricular late potentials, Meth. Inform. Med., 33(2), 187–195, 1994.

34. Meste, O., Rix, H., Caminal, P., and Thakor, N.V., Ventricular late potentialscharacterization in time–frequency domain by means of a wavelet transform,IEEE Trans. Biomed. Eng., 41, 625–634, 1994.

35. Taboada-Crispi, A., Improving Ventricular Late Potentials Detection Effective-ness, Ph.D. thesis, Univeristy of New Brunswick, 2002.

36. Schiff, S.J., Aldroubi, A., Unser, M., and Sato, S., Fast wavelet transformationof EEG, Electroencephalogr. Clin. Neurophysiol., 91(6), 442–455, 1994.

37. Kalayci, T. and Ozdamar, O., Wavelet preprocessing for automated neuralnetwork detection of spikes, IEEE Eng. Med. Biol., 14(2), 160–166, 1995.

38. Blinowska, K.J., and Durka, P.J., The application of wavelet transform andmatching pursuit ot the time-varying EEG signals, in Intelligent EngineeringSystems Through Artificial Neural Networks, Vol. 4, Dagli, Fernandez, and Gosh,Eds., 1994, pp. 535–540.

39. Donoho, D.L., De-noising by soft thesholding, IEEE Trans. Inform. Theory, 41,613–627, 1995.

40. Lewis, A.S. and Knowles, G., Image compression using the 2-D wavelettransform, IEEE Trans. Image Process., 1, 244–250, 1992.

41. Angelidis, P.A., MR image compression using a wavelet transform codingalgorithm, Magnetic Resonance Imaging, 1(7), 1111–1120, 1994.

42. Hilton, M., Wavelet and wavelet packet compression of electrocardiograms,IEEE Trans. Biomed. Eng., 44, 394–402, 1997.

43. Shapiro, J.M., Embedded image coding using zerotrees of wavelet coefficients,IEEE Trans. Signal Process., 41(12), 3445–3462, 1993.

44. Wellig, P., Cheng, Z., Semling, M. and Moschytz, G.S., Electromyogram DataCompression Using Single-Tree and Modified Zero-Tree Wavelet Encoding,Proc. 20th Int. Conf. IEEE Eng. Med. Biol. Soc., 20(3), 1303–1306, 1998.

45. Norris, J., Englehart, K., and Lovely, D.F., Myoelectric signal compressionusing the EZW algorithm, Conf. of the IEEE Engineering in Medicine and BiologySociety, Istanbul, October, 2001.

46. Carmona, R. and Hudgins, L., Wavelet denoising of EEG signals and identi-fication of evoked response potentials, Proc. SPIE Conf. Wavelet Appl. SignalImage Process., 2303, 91–104, 1994.

47. Tikkanen, P.E,. Nonlinear wavelet and wavelet packet denoising of ECGsignal, Biol. Cybernetics, 80(4), 259–267, 1999.

48. Bentley, P.M., Grant, P.M., and McDonnell, J.T.E., Time-frequency and time-scale techniques for the classification of bioprosthetic valve sounds, IEEETrans. Biomed. Eng., 45(1), 1998.

49. Mallat, S. and Zhong, S., Characterization of signals from multiscale edges,IEEE Trans. Pattern Recog. Machine Intell., 14(7), 1992.



50. Senhadji, L., Carrault, G, Ballanger, J.J., and Passariello, G., Comparing wave-let transforms for recognizing cardiac patterns, IEEE Eng. Med. Biol.,March/April, 167–172, 1995.

51. Li, C., Zheng, C., and Changfeng, T., Detection of ECG characteristic pointsusing wavelet transforms, IEEE Trans. Biomed. Eng., 42(1), 1992.

52. Lim, L.M., Akay, M., and Daubenspeck, A.J., Identifying respiratory–relatedevoked potentials, IEEE Eng. Med. Biol., March/April, 174–178, 1995.

53. Bovik, A.C., Gopal, N., Emmoth, T., and Restrepo, A., Localized measurementof emergent image frequencies by Gabor wavelets, special issue on wavelettransforms and multiresolution signal analysis, IEEE Trans. Inform. Theory,IT–38(3), 691–712, 1992.

54. Healy, Jr., D.M. and Weaver, J.B., Two applications of the wavelet transformin magnetic resonance imaging, IEEE Trans. Inform. Theory, 38, 840–860, 1992.

55. Englehart, K., Hudgins, B., and Parker, P.A, A Wavelet Based ContinuousClassification Scheme for Multifunction Myoelectric Control, IEEE Trans.Biomed. Eng., 48(3), 302–311, 2001.

56. Englehart, K., Hudgins, B., Parker, P.A., and Stevenson, M., Classification ofthe myoelectric signal using time-frequency based representations, specialissue on intelligent data analysis in electromyography and electroneurogra-phy, Med. Eng. Phys., 21, 431–438, 1999.

57. Thakor, N.V., Gou, X., Sun, Y.–C., and Hanley, D.F., Multiresolution waveletanalysis of evoked potentials, IEEE Trans. Biomed. Eng., 1085–1094, 1995.

58. Kadambe, S. and Boudreaux–Bartels, G.F., Applications of the wavelet trans-form for pitch detection of speech signals, IEEE Trans. Inform. Theory, 38,917–924, 1992.

59. Dorcas, D. and Scott, R.N., A three state myoelectric control, Med. Biol. Eng.,4, 367–372, 1966.

60. Childress, D.A., A myoelectric three state controller using rate sensitivity, inProc. 8th ICMBE, Chicago, IL, 1969, pp. S4–S5.

61. Vodovnik, L., Kreifeldt, J., Caldwell, R., Green, L., Silgalis, E., and Craig, P.,Some Topics on Myoelectric Control of Orthotic/Prosthetic Systems, Report No.EDC 4–67–17, Case Western Reserve University, Cleveland, OH, 1967.

62. Hudgins, B., Parker, P.A., and Scott, R.N., A new strategy for multifunctionmyoelectric control, IEEE Trans. Biomed. Eng., 40(1), 82–94, 1993.

63. Hudgins, B., Englehart, K., Parker, P.A., and Scott, R.N., A microprocessor-based multifunction myoelectric control system, in Proc. 23rd Canadian Medicaland Biological Engineering Society Conf., Toronto, Canada, May, 1997.

64. Englehart, K., Hudgins, B., Stevenson, M., and Parker, P.A., Classification oftransient myoelectric signals using a dynamic feedforward neural network,World Congress of Neural Networks, Washington, D.C., 1995.

65. Farry, K.A., Fernandez, J.J., Abramczyk, R., Novy, M., and Atkins, D., Apply-ing genetic programming to control of an artificial arm, in Proc. MyoelectricControl '97 (MEC ’97) Conference, Fredericton, New Brunswick, Canada, July23–25, 1997, pp. 50–55.

66. Leowinata, S., Hudgins, B, and Parker, P.A., A multifunction myoelectriccontrol strategy using an array of electrodes, in Proc. 16th Annual Congress ofthe International Society Electrophysiology and Kinesiology, Montreal, Canada,1998.



67. Gallant, P.J., An Approach to Myoelectric Control Using a Self–OrganizingNeural Network for Feature Extraction, Master’s thesis, Queens University,Kingston, Ontario, 1993.

68. Coifman, R.R. and Wickerhauser, M.V., Entropy-based algorithms for bestbasis selection, IEEE. Trans. Inform. Theory, 38(2), 713–719, 1992.

69. Nason, G.P. and Silverman, B.W., The stationary wavelet transform and somestatistical applications, Lect. Notes Statistics, 103, 281–299, 1995.

70. Rioul, O. and Vetterli, M., Wavelets and signal processing, IEEE Signal Process.Mag., 8, 14–38, 1991.

71. Shensa, M.J., Affine wavelets: wedding the à trous algorithm and Mallat lags,IEEE Trans. Signal Proces., 40, 2464–2482, 1992.

72. Cohen, A., Daubechies, I., and Feauveau, J.C., Biorthoginal bases of compactlysupported wavelets, Communications of Pure and Applied Math, 1990.

73. Pesquet, J.C., Krim, H., and Carfantan, H., Time invariant orthonormal wave-let representations, IEEE Trans. Signal Processing, 44(8), 1964–1970, 1996.

74. Cohen, I., Raz, S., and Malah, D., Shift invariant wavelet packet bases, Proc.of ICASSP, 2, 1081–1084, 1995.

75. Coifman, R.R. and Donoho, D.L., Translation-invariant de-noising, in Waveletsand Statistics, Antoniadis, A. and Oppenheim, G., Eds., Lecture Notes inStatistics, Springer-Verlag, Berlin, 1995, pp. 125–150.

76. DelMarco, S. and Weiss, J., Improved transient signal detection using a wave-packet-based detector with an extended translation-invariant wavelet trans-form, Optical Eng., 35(1), 131–137, 1996.

77. Liang, J., and Parks, T.W., A translation–invariant wavelet representationalgorithm with application, IEEE Trans. Signal Process., 44(2), 225–232, 1996.

78. Simoncelli, E.P., Freeman, W.T., Adelson, E.H., and Heeger, D.J., Shiftablemultiscale transforms, IEEE Trans. Inform. Theory, 38(2), 587–607, 1992.

79. Nason, G.P. and Silverman, B.W., The stationary wavelet transform and sta-tistical applications, in Wavelets and Statistics, Antoniadis, A. and Oppenheim,G., Eds., Lecture Notes in Statistics, Springer-Verlag, Berlin, 1995, pp. 281–299.

80. Mallat, S. and Zhang, Z., Matching pursuit with time-frequency dictionaries,IEEE Trans. Signal Process., 41, 3397–3415, 1993.



chapter nine

Neuroprosthetic device design

Donald L. Russell

Contents

9.1 Introduction9.2 The device classes

9.2.1 Devices with neural input9.2.2 Devices with neural output9.2.3 Devices with both neural input and neural output

9.3 The design challenges9.3.1 Signal measurement9.3.2 Signal interpretation9.3.3 Intent interpretation9.3.4 Performing the desired action9.3.5 Signal generation9.3.6 Signal transmission

9.4 An example: prosthetic arms9.4.1 State-of-the-art prosthetic elbows9.4.2 Signal measurement and interpretation9.4.3 Intent interpretation9.4.4 Performing the desired action9.4.5 Signal generation and transmission

9.5 Other neuroprosthetic devices9.5.1 Cochlear implants9.5.2 Bladder

9.6 ConclusionsReferences



9.1 IntroductionThe goal of this chapter is to provide the researcher with a general contextfor the design of neuroprosthetic devices. Specifically, the practical issuessurrounding the design and development of devices that will actually beused are presented. Concepts are given and exemplified by referring to thedevelopment of powered prosthetic elbows. In short, this chapter attemptsto describe the overall design challenges (the “forest”) rather than the detailsof any specific area (the “trees”).

Neuroprosthetic devices are objects that interact with the nervous systemto perform a function that maintains or improves the health and quality of lifeof an individual. They can be broken down into three classes: those that makeuse of neural measurements as an input to control a mechanical device thatreplaces a function of the body (for example, a prosthetic hand that uses myo-electric based control); those that, based on some action or situation, create anoutput signal that is transmitted to a nerve to have a desired action (for example,neuroprosthetic bladders); and those that have both neural inputs and neuraloutputs (for example, devices aimed at bypassing spinal cord damage).

Even when neural signals can be reliably measured, interpreted, andtransmitted, a device must be constructed to perform the required function.The technical demands on such devices may be significant, as they typicallyrequire their own power sources and cannot take advantage of the abilityof the body to repair damage and compensate for wear. Neuroprostheticorgans that process material may require supplies of material and the abilityto discharge waste material. Further, internal devices must work within stricttemperature limits or else surrounding tissues will be destroyed.

To be useful (in the eyes of the user) the device must improve both healthand quality of life, it must be reliable, the performance of the device mustbe predictable, the device must be safe and not create other health hazardsas a result of its operation, and the effort and concentration required intraining and day-to-day use must be reasonable. These are significant chal-lenges. The use of the device will be compared to the ability of the user to“make-do” and adapt to life without using the device. The ability of peopleto adapt and find innovative solutions to challenging situations after the lossof function of a limb or organ is significant. This is, however, the level offunction that must be surpassed if the device is to be used. Such mundanefeatures as the appearance, operating life, or time between recharging bat-teries of the device often have a significant impact on the very practicaldecision that a user makes about using a device.

This chapter will present a general classification of different types ofdevices and the challenges faced in their practical design. A review is pre-sented of current approaches to the design of a number of current neuro-prosthetic devices: limbs, the bladder, and the ear. Finally, the impact ofcurrent technologies on the design of practical devices is briefly discussed.The primary example used in this chapter will be the powered, myoelectri-cally controlled prosthetic elbow.



9.2 The device classes9.2.1 Devices with neural input

This class of neuroprosthetic devices consists of those that base their oper-ation, either directly or indirectly, on measurements of neural activity (seeFigure 9.1). The measurements are then interpreted to create control signalsappropriate for the device. Those signals are then used to control the actionof the device, be it movement, the release of materials, or some other action.In most cases, some sort of feedback exists. This feedback is critical for thebody or device to regulate its action. In some instances the feedback may beentirely confined to the device; in other cases, the body’s natural senses (suchas vision) and nervous system may provide the feedback.

9.2.2 Devices with neural output

In this class of devices the stimulus for action does not arise from measure-ment (direct or indirect) of neural activity but from some other action (forexample, pushing a button). This input signal is generally processed andmanipulated to create a signal that is designed for transmission to a nerve(see Figure 9.2). Because this signal is to be transmitted back into a nerve,the design must take into account knowledge of other associated neuralconnections. Devices, such as pacemakers, which respond to the needs ofthe body but not to direct measurement of neural signals, fall into thiscategory as they generate signals that interact with the nervous system. Agroup of devices that is placed in this class (rather than the second class) arethose that are used to implement functional neuromuscular stimulation.These devices measure a nervous signal or some other input signal andtransmit this signal, via electrodes, to a muscle. In this way external devicescan control aspects of the body over which the subject has lost control.

Figure 9.1 Interactions with a neuroprosthetic device that uses measurement of neu-ral signals as a source of control information.

Measurementand

Interpretation

NeuroprostheticDevice

NerveSignal Control

Signal

PowerandMaterials

TheBody

Action, Movement,work, etc.

OtherSignals

Feedback to Body

Feedbackto Device



9.2.3 Devices with both neural input and neural output

This third class of device essentially replaces the transmission function ofthe nervous system or modifies a nerve signal to achieve desired goals(Figure 9.3). Both its output and input are neural signals so that the chal-lenges of accurate interpretation of a measured nerve signal and of creatingan appropriate, interpretable output signal are significant. In general, devicesin this category fulfill the role of transmission of the nervous signal and assuch do not have significant material or power requirements (relative todevices that actually move or manipulate material).

Figure 9.2 Interactions with a neuroprosthetic device that creates a signal that istransmitted to a nerve to achieve a desired function.

Figure 9.3 Interactions with a neuroprosthetic device that measures neural activityand transmits it (or a modification of it) back to the nervous system.

Interpretation


Nerve Signal

ControlSignal

PowerTheBody

Action,movement,work,etc.

Feedbackto Body

CommandSignal


Nerve Signal

ControlSignal

Power

TheBody

Action,movement,work,etc.

Feedbackto Body

Measurementand

Interpretation

NerveSignal

OtherSignals



9.3 The design challengesThe challenges in designing practical neuroprosthetic devices are many. Theygenerally include signal measurement, signal interpretation, performing thedesired action, intent interpretation, signal generation, and signal transmis-sion. These challenges are outlined here from the point of view of thedesigner of a practical device. A specific device may require considerationof some or all of these challenges.

9.3.1 Signal measurement

For a device to be useful and to be used it must be reliable. The ability tomeasure a signal from neural inputs over a span of time without the needfor significant medical intervention is critical. Components that gather thesignals must do so in a way that is not subject to a response from the bodythat inhibits their function, either directly or through a side effect. Conquer-ing these challenges requires a range of efforts and solutions, including thecreation of implantable electrodes that use materials and designs that avoidthese problems, the careful placement of surface electrodes to avoid skinabrasion or local pain, and the development of other sensors to measurevarious quantities the represent the state of the body.

9.3.2 Signal interpretation

The neural signal, once measured, must be interpreted; the designer musthave the answer to the question, “What does the signal mean?” The body’snetwork of nerves is extremely complex, offering many opportunities forinteraction of neural signals with related (or sometimes unrelated) actionsand significant feedback (usually to the central nervous system). This net-work is also nonlinear and the signals measured may not be directly inter-pretable in terms of the quantities and variables engineers and scientists maychoose to describe the resulting actions. The parts of the body that themeasured signals would normally control and interact with are, in general,far more complex and nonlinear than the devices used to replace them. Inmany cases, the state of the art of a device represents a first-order solutionto a highly complex problem.

Artificial neural networks often can achieve high levels of robust per-formance, yet the internal signals within those networks are not meaningful.In most cases where neural measurement is used to control a device, themeasured signals are those from within the network of nerves. By analogywith artificial neural networks, these signals may not have any apparent orsimple meaning. Significant amounts of experimentation on the system con-cerned must be performed in order to arrive at an interpretation of a neuralsignal that can be used to infer a desired action (for example, to flex thelimb). This correlation between a measured neural signal and the intendedaction represents a significant challenge, especially as the details almost



certainly vary from individual to individual and because injury or diseasemay alter “normal” patterns of activity.

9.3.3 Intent interpretation

This challenge is analogous to the challenge of signal interpretation butapplies to inputs that indicate the state of the body and not the actionintended by the body. For example, future artificial hearts may respond tomeasurements of oxygen or carbon dioxide levels within the blood or aninput from a physician indicating a lack of oxygen in some organ. Havingsuch an input then requires the decisions to be made – what should be done?In this example, should the heart rate be increased or lowered? Are otheraspects of the body that are unmeasured (such as, the presence of internalbleeding) important in making this decision? The body is a complex systemwith many possible solutions to a given problem but also with many possibleoutcomes to a single action. Designers must focus on the whole body andin creating responses to specific problems.

9.3.4 Performing the desired action

Once an action is required, the device itself must be able to adequatelyperform the action. In the case of prosthetic limbs, significant energy isrequired to manipulate objects. Power and material are often required. Pace-maker batteries last for years and performance of the pacemaker itselfdegrades “gracefully” as power levels drop. Replacement of the batteryrequires minor surgery but this is infrequent enough that it is not a significantdeterrent to the use of pacemakers. Prosthetic limbs also work on batteries.In this application, however, limbs must be designed with removable batterypacks so that a single day of practical use is possible. This has a significantimpact on limb design. Challenges surrounding the realistic performance ofa device are often those that limit the development of practical devices. Theamazing capabilities of the body make the challenges of designing replace-ment parts significant.

9.3.5 Signal generation

When a neural signal is to be created, enough knowledge of the system mustexist to understand what signal is required to achieve a given goal or require-ment. This is the inverse of the signal interpretation problem. A designermay be able to create an input that indicates a desired action. However,inserting a neural signal into the midst of a complex array of neuronsdepends on knowing the body’s neural representation corresponding to thedesired action and a careful and complete understanding of other possibleresults of the neural signal. Examples are the muscle fatigue and musclecramps that can result when the output signal of a functional neuromuscularcontroller is not carefully controlled. If the neuroprosthetic device functions



to gather sensory information, a large quantity of detailed information mayhave to be processed and transmitted to nerves.

9.3.6 Signal transmission

Even when a signal can be created that will do the appropriate thing withoutinducing other problems, the practical issues of transmitting the signal backto a nerve with an approach that is reliable and robust over the long termcan be challenging. Other more practical issues often can have a great impacton the usefulness of the neuroprosthetic devices. For example, the devicesmust by mounted and supported so that their operation is not affected bymovement of the user. The techniques used to locate the device must notcause irritation or other problems for the user.

9.4 An example: prosthetic armsA limb is a complex set of tissues. Limbs are the primary way in which weinteract with the physical world. The loss of a limb through disease oraccident is a traumatic occurrence for which physicians and engineers havefew solutions. In terms of the variety and complexity of tasks few neuro-prosthetic devices offer as many challenges to be overcome. One of thesimplest joints in the human body is the elbow and, because of this, thepowered prosthetic elbow will be used to illustrate examples of the types ofchallenges outlined in the previous section.

The human arm has many uses, and an arm amputee using a neuro-prosthetic device has an expectation that the prosthesis will be capable ofperformance approaching that of the human arm. In fact, discussions withamputees show that many amputees have expectations that their roboticarms will be much more powerful based on various science-fiction stories.This expectation that an artificial device should be in some way better thanthe biological tissues it attempts to replace is a non-technical challenge facedby all those who assist patients in the training and use of these devices.

9.4.1 State-of-the-art prosthetic elbows

The ideal prosthetic arm would replace all of the functions of the natural arm.It must be strong, but also gentle, and fast, but also capable of precision, itmust be able to operate with a range of stiffness. It must be able to sense allof those senses that the arm possesses (e.g., heat, pressure); it must look andmove in a manner that gives the appearance of the natural limb. Further, thearm must respond correctly and naturally to the neural commands that aresent to it. All sensing capabilities of the arm must be translated back into neuralsignals and transmitted back into the body. These are challenging requirementsthat are discussed in more detail below. A complete historical review of pros-thetic arms is presented in Mann.1 Two of the more common, commerciallyavailable neuroprosthetic limbs are the Boston arm2 and the Utah arm.3



The performance of commercial prosthetic elbows does not approachthat of the intact arm. They are capable of lifting a maximum load that istypically much less than 10 lb; for example, the Boston elbow can lift 9 lb atthe hand.4 While most prosthetic elbows are fitted with some type of mechan-ical clutch to hold static loads, these clutches do not mimic natural behavior.Elbow speed varies slightly with the limb’s orientation relative to gravity,but typical limbs move with a maximum movement time from one limit ofmotion to the other (roughly 135 degrees) of about 1 sec; for example, theUtah arm will flex against gravity in 1.1 sec.4

Several limbs also incorporate other modes of movement. For example,keeping the limb rigid during walking causes the amputee to appear unnat-ural. In order to overcome this, the Utah arm incorporated a powered “free-swing” mode, and the Boston arm incorporates a mechanical “free-swing”mode. The goal of these modes is to allow the arm to swing naturally duringwalking. In order to change the behavior of the limb from a movement-controlled mode to a free-swing mode (or any other mode), some sort ofswitching is required. This can be accomplished by the use of switches orby the identification of special features in the neural inputs (for example,the Utah arm uses a rapid simultaneous contraction of two neural inputs toswitch between modes).

Commercial powered limbs work on removable, rechargeable batteries.These batteries must be removable because they are not capable of providingthe limb with enough energy for a full day of use. Amputees requiring aprosthetic elbow also require terminal devices to allow them to interact withenvironment. While beyond the scope of this discussion, the need to controlfurther degrees of freedom, such as a hook or hand, requires either a greaternumber of operation modes (and the complexity of switching between them)or more inputs.

9.4.2 Signal measurement and interpretation

The human arm responds to a large number of nerves that can control awide variety of movements. Even restricting the discussion to prostheticelbows, many muscles cross the elbow joint and each muscle has many motorunits which are controlled by different neural inputs. Questions and hypoth-eses about how these neural commands work together to produce the coor-dinated, smooth motions of a limb abound and present many challenges.5Variations from individual to individual make data analysis difficult. Possi-ble changes in neural behavior as a result of the amputation or perhaps thebody’s adaptation to the amputation further complicate the need to under-stand the relation between neural signal and intended action.

Neuroprosthetic limb designers have developed a range of solutions tothe challenge of measuring signals that can be used to control the limb. Ingeneral, commercial limbs use measurements of myoelectric activity byusing surface electrodes. These electrodes (typically mounted inside thesocket of the limb so that in use they sit on the surface of the stump over



appropriate muscles) sense the neural signals as they are transmittedthroughout a muscle or group of muscles. The detected signals containsuperimposed information from active motor units of any muscles closeenough to the electrode to make their signal measurable. All of the signalsare further modified due to the nature of transmission through the hetero-geneous material in the arm between the active muscles and the electrodes.The raw signals measured in this way do not appear to differ significantlyfrom white noise. The use of surface measurement of these signals hasproven to be a practical, reliable, and robust technique of gaining informationon the neural input to the limb.

Other inputs are possible, including the use of cables arranged so thatrounding of the shoulders will flex the elbow, the use of switches that convertsome other body movement into a command for the elbow, and the use ofsignals measured from other muscles. The typical drawback encounteredwith these inputs that couple another action of the body to the motion ofthe elbow is the loss of a degree of freedom; however, directly coupling themotion of a part of the body to the motion of the limb can allow a kind offeedback. This approach, incorporating extended physiological propriocep-tion, has a number of benefits and has been a subject of recent research.6

Two basic approaches exist to extract useful information from this formof neural activity measurement. The differences in approach reflect deci-sions made about the level of physiological meaning assumed in use ofthe measured signals. The first approach is fairly commonly used, withvariations in the details of implementation, in most myoelectric prostheticsituations.2,3 Signals are measured using surface electrodes. Typically, thesignals are processed by first rectifying the signal and then calculating amean squared amplitude estimate. No attempt is made in this type ofapproach to differentiate between different muscles or motor units thatmay be contributing to the myoelectric signal. The second approach usesan array of electrodes to collect information on the spatial distribution ofthe signal over the surface of the arm.7 This information is processed todetermine more accurate estimates of the signals in the muscles. Thissecond approach requires significantly more computational effort and isfar more complex than the first approach.

The use of surface electrodes also can cause problems ranging from skinirritation with prolonged use to the presence of extraneous components inthe measured signals (such as the so-called movement artifacts, which resultfrom the movement of the electrode as the muscle contracts and changes thelimb geometry).

9.4.3 Intent interpretation

After the measurements are taken they must be interpreted. A significantamount of effort has been, and continues to be, put into establishing themeaning of these signals. In fact it is not necessary to be able to determinethe physiological meaning or the intent behind the signals in order to use



them to control a prosthesis. Several prosthetic control schemes8 base theaction of the prosthesis on the magnitude of a single measured signal. Forexample, a low signal may correspond to a closed hand, and a higher signalmay cause a hand to open. In the case of a joint, a low signal below anestablished threshold would cause the arm to remain stationary, a signalabove a certain higher established threshold may cause the arm to flex, anda signal between the two thresholds would cause the arm to extend. Byextrapolation, higher levels of discretization of the magnitude can achievemore actions. The limit on this control approach is the amputee’s ability togain a facility at precisely regulating the signal amplitude. This is compli-cated by the lack of any close relationship between the original meaning ofthe signal and its use. Also, amputees often have difficulty regulating thesesignals without much practice and focused concentration. These difficultieslimit the number of levels of signal magnitude that can be differentiated.Recent work has identified characteristics of the myoelectric signal duringthe onset of activity that can be used to switch between different devices orcontrol actions.9

The measured signal can be interpreted in other ways. One of thecommon ways of interpreting the signal relies on having signals from twomuscles in an agonist–antagonist configuration. For example, in the caseof an above-elbow amputation, signals from the biceps muscle remnantand from a triceps muscle remnant can be used together to infer somethingabout the intended movement of the limb. In particular, it is commonpractice to first rectify and average the individual signals and then use thedifference between the signal amplitudes as the control input. Specifically,a large difference in the signals implies a large joint velocity, and a smalldifference implies a small joint velocity. Care is taken to ensure that a “deadzone” exists so that neural activity below a certain threshold does not resultin any motion.

While this approach is more physiological than the switching control,in reality the relationship between measured neural or myoelectric activityand limb movement is far more complex. It also results in limb motion thatappears to be very robotic. That is, prostheses using this control schememove in a manner that is easily distinguishable from the movement of ahealthy limb. As will be discussed more below, appearance is a critical factorin designing a prosthetic limb that will be used. In fact, many prostheticarms now incorporate separate modes of operation (e.g., the free-swingmode in the Utah arm) that ensure that the limb has a more natural appear-ance during walking. In order to detect when a switch between modes ofoperation (controlled movement and free swing) is desired, sudden co-con-traction events are used. When the amputee suddenly contracts both mus-cles, the prosthesis will switch modes.

Research efforts are extending the approach of using two antagonisticmuscles to generate control inputs. Efforts by Hogan10 suggest using the sumof the two signals as an indication of limb stiffness, thus mimicking thenatural arm in which co-activation of antagonist muscles results in a stiffness



increase of the limb. Preliminary experiments on a prosthesis emulator usedfor the evaluation of new control schemes indicated significant promise inthis approach. However, significant changes in limb design are necessarybefore such an approach can be used in a practical commercial limb.

9.4.4 Performing the desired action

If the intent of the user is known the limb must be able to execute thismovement. Knowing what to do is only part of the problem. In many cases,the significant challenges are actually accomplishing the desired tasks. Inthe case of the prosthetic elbow, determining what the desired action shouldbe is a significant challenge but most commercial limbs are severely limitedin their mechanical performance.

An actual limb has very complex mechanical properties. Commerciallimbs respond to inputs by moving at specific velocities, without regard forany contact between the limb and the environment. Muscles, as actuators,have very different properties from electric motors and mechanical trans-missions. One of the fundamental differences between these actuators is thestiffness (or more generally, the impedance), the response of the actuatorwhen it is forced to move from its commanded equilibrium position.

Muscles have relatively low stiffness when compared to motors and gearsets.11 Typical robotic control schemes have the goal of ensuring that the robotis moving as commanded. Many commercial prosthetic controllers share thisfeature. Designers must consider what would happen if the limb is forced tomove by an external force, as might occur when bumping an object or inattempting to carry an object for which the weight exceeds the lift capacityof the arm. Only two possibilities are possible: either the motor and trans-mission prevent the motion from occurring or the device is designed so thatit is possible for an external force to move the limb. If the motion is prevented,both the limb and the amputee must support the applied load. As impactloads can be significant this would generally result in damage to the pros-thesis and pain for the amputee as the load is transmitted from the prosthesisto the residual limb. This is why commercial limbs will move when a sufficientload is applied to them (they are “back-drivable”).

For the second possibility, consider a limb that is moving as directed, atthe speed corresponding to the measured muscle signals, but bumps into anobject. The object may be moved or damaged. More commonly, the objectis too heavy to be moved by a prosthetic limb and too strong to be damagedby it. The result is that the controller causes the arm to apply as much forceas it can to achieve the desired movement. When that force surpasses themaximum force that can be generated by the motor, the limb is forced tomove from its commanded position, overpowering the motors and creatinga high-pitched squealing sound. As one of the main goals is to provide theamputee with the ability to participate in society and to feel acceptable, thishigh-pitched sound is a significant drawback for many users as it bringsattention to their missing limb.



The general idea of the appearance (or cosmesis) of the limb is significantto amputees.12 Approximately 75% of arm amputees who choose not to weara prosthesis have looked for a device that appeared the same as their soundarm.13 For many individuals, the need to have a limb that looks like it is real,both when stationary and when in motion, is one of the major factors dictatingthe level of use of the prosthesis. Also of significance is the manner in whichthe amputee supports the limb; the fit of the socket is critical for the long-term comfort and to prevent irritation of the residual limb during use.

One feature of the way that the healthy body responds to neuralcontrols is the lack of a requirement for concentration. For most people,activities such as listening, keeping the heart beating, or using a limb towalk or move a sandwich from the plate to the mouth do not requiresignificant focus and concentration. Amputees learning to use a new pros-thetic limb can require very significant levels of concentrated mental effort.Training periods that are too long with too little indication of progress anda return to normal function can reduce the morale and interest of theamputee to a point where interest in gaining a facility in using the devicedisappears. The most advanced neuroprosthetic limbs will be unused ifthe training requirement is too long and difficult. Each individual user mayhave a different level of tolerance for extensive training and a differentlevel of desire to master the device.

Finally, for many prosthetic devices power use is an issue. Pacemakersuse power at a rate that allows them to function for years before new batteriesare needed. However, for prosthetic limbs significant power is necessary; infact, one view of the interaction of a limb and the world is to describe it asthe controlled distribution and absorption of kinetic energy. An arm providesenergy to an object as its velocity increases and absorbs energy as its velocityis reduced. The need for removable and/or rechargeable batteries in pros-thetic arms has a significant impact on the geometry of the limb and theefficiency of space usage. An increase in the efficiency of the limb or its usewill lengthen the time before it is necessary to replace or recharge the bat-teries. When a design is achieved that allows limbs to be used for a completeday of use, many other aspects of the designs can change. Under suchconditions, batteries could be distributed in the device and charging occurovernight when the limb is not in use.

Given the limited performance of current technology, many advances inartificial limbs are possible. One promising approach is to construct theprosthetic limb using biomimetic principles. That is, the limb can be designedso that its mechanics mimic an intact limb. With this approach, muscle signalsmight be fed directly to actuators without the need to interpret the signal orperform extensive manipulations on the signal.

9.4.5 Signal generation and transmission

Commercial prosthetic limbs do not provide direct feedback to the user toreplace the sensations that were present in the amputated limb. The volume



of sensory information to be transmitted is such that researchers concentrateon provision of very limited sensing back to the user.

9.5 Other neuroprosthetic devicesWhile entire books or chapters could be written on each of these devices,this section summarizes the current state of a number of these devices withreference to the design challenges given above. Specifically, the ear (cochlearimplant) and the bladder are parts of the body for which commercial neu-roprosthetic devices have been developed. These two areas are discussedbriefly here as examples of where the technology can lead. The large bodyof work on pacemakers and similar cardiovascular devices that interact withthe nervous tissue surrounding the heart is not discussed here.


These neuroprosthetic devices are intended to provide hearing to the deaf.14

They function by converting sounds picked up by a microphone into signalsthat are transmitted with an array of small, implanted electrodes to theauditory nerve. A microphone worn by the user collects sound informationthat is processed, commonly with digital signal processing (DSP)-basedhardware to process and extract features from the measured sound. A signalis then created that is transmitted to the implanted portion of the device(often using radiofrequency [RF] technology). The signal is then used tostimulate the auditory nerve with electrodes. These electrodes must be care-fully placed to generate useful signals in the auditory nerves. While no claimis made that hearing is restored, companies such as Advanced Bionics(www.bionicear.com) and AllHear, Inc. (www.allhear.com) suggest that theirdevices provide enough sensation and information about the sound envi-ronment for many to recognize speech, especially in conjunction with expe-rience in other techniques such as lip reading.

The challenges confronting the design or design improvements tocochlear implants follow the general outline presented here. Collecting theinput signal (sound) using a microphone should be done in a way that allowsthe user to participate in society without significant, visible signs of the useof the device. Perhaps the biggest challenge in the design of these neuro-prosthetic devices is dealing with the large volume of information in themeasured sound. Current solutions to the problem involve extracting fea-tures from the sound or processing the sound in a fashion that reduces theinformation content to a level that can be transmitted into the nervoussystem. By improving the level of understanding of the way that the healthyear functions these solutions can be improved so that a larger amount ofmore accurate information can be transmitted to the auditory nerves.

The components that process the signals and generate the signal to betransmitted to the implants are external units that must be carried by theuser. Further development of signal processing electronics to reduce the size



http://www.bionicear.com/

http://www.allhear.com/

and power requirements of this part of the technology will provide a farmore user-friendly product that is less intrusive into other aspects of theusers life. Even when the signal is properly generated, it must be transmittedto the nervous system. Careful development of finer arrays of electrodes (orthe equivalent) and development of techniques for implanting, locating, andcalibrating these electrodes are two other important steps in moving thesedevices forward.

9.5.2 Bladder

The loss of bladder control presents an individual with a medical problemthat also has serious social implications. Regaining control of the bladderallows these individuals to improve their health and participate and con-tribute to society without fear of embarrassment. Commercial neuropros-thetic bladders (for example, the Vocare system made by Neurocontrol Corp.;remote-ability.com) provide the patient with an external unit to control whenthe bladder will empty. These units generate signals that are transmitted toan implanted receiver, which uses pacemaker-like technology to activateappropriate nerves. Future challenges in neuroprosthetic bladder develop-ment center on device design to make the devices smaller and less obtrusiveand to perhaps develop approaches that allow the user control of the devicewithout the need for an external triggering unit.

9.6 ConclusionsThe design of actual, usable neuroprosthetic devices requires an understand-ing of the complex details underlying the function of the body the device isintended to replace. For the devices that are developed to be practical andused by the client group they were designed for, a holistic approach mustbe taken to their design. The chapter has outlined the basic nature of thechallenges facing neuroprosthetic device designers. Ideally, the devices mustbe able to perform the required actions in a way that does not impact theusers’ health in another way. Muscles and nerves have fundamentally thesame principles of operation as their technological equivalents — motorsand wires.

References1. Mann, R.W., Cybernetic limb prosthesis: the ALZA distinguished lecture,

Ann. Biomed. Eng., 9, 1–43, 1981.2. Williams, III, T.W., The Boston elbow: the path to a mature myoelectric pros-

thesis, SOMA, 29–32, 1986.3. Jacobsen, S.C., Knutti, D., Johnson, R., and Sears, H., Development of the

Utah artificial arm, IEEE Trans. Biomed. Eng., BME-29(4), 249–269, 1982.4. Liberty Technology, Prosthetics and Orthotics Group, The Liberty Boston Elbow:

A Complete Prosthetic Arm System, Liberty Technology, Hopkinton, MA, 1997.



5. Winters, J.M. and Woo, S.L.-Y., Eds., Multiple Muscle Systems: Biomechanics andMovement Organization, Springer-Verlag, New York, 1990.

6. Richard, F., Heckathorne, C.W., and Childress, D.S., Cineplasty as a controlinput for externally powered prosthetic components, J. Rehab. Res. Dev., 38(4),2001.

7. Clancy, T.E., Morin, E.L., and Merletti, R., Sampling, noise-reduction andamplitude estimation issues in surface electromyography, J. Electromyogr. Ki-nesiol., 12(1), 2002.

8. Scott, R.N., Parker, P.A., O’Neill, P.A., and Morin, E.L., Criteria for settingswitching levels in myoelectric prostheses, J. Assoc. Children’s Prosthetic-Orthotic Clin., 25(1), 1990.

9. Hudgins, B., Englehart, K., Parker P., and Scott, R.N., A microcomputer-basedmultifunction myoelectric control system, CMBEC ’97, Toronto, 1997.

10. Abul-Haj, C.J. and Hogan, N., Functional assessment of control systems forcybernetic elbow prostheses. I: Description of the technique, II: Applicationof the Technique, IEEE Trans. Biomech. Eng., 37(11), 1990.

11. Burke, T., Modelling and Evaluation of Nonbackdrivable Transmissions forVariable Stiffness Prostheses, Master’s thesis, Carleton University, Ottawa,Canada, 2000.

12. Kaczkowski, M. and Jeffries, G.E., Cosmesis is much more than appearance… it’s function, Motion, 9(3), 1999.

13. Melendez, D. and Leblanc, M., Survey of arm amputees not wearing pros-theses: implications for research and service, J. Assoc. Children’s Prosthetic-Orthotic Clin., 23(3), 1998.

14. NIH, Cochlear Implants, Health Information Bulletin, NIH Doc. No. 00–4798,National Institutes of Health, Washington, D.C., 2000.




section five

Prosthetic systems


chapter ten

Implantable electronic otologic devices for hearing rehabilitation

Kenneth J. Dormer

Contents

10.1 Introduction10.2 Review of auditory mechanisms10.3 Hearing losses10.4 Types of implantable hearing devices

10.4.1 Bone conduction devices10.4.2 Middle ear devices10.4.3 Cochlear implants

10.5 Signal processing10.5.1 Cochlear electrode types10.5.2 Types of cochlear stimulation10.5.3 Auditory brainstem implants

10.6 SummaryReferences

10.1 IntroductionBeginning in the 1960s, interest was generated among auditory physiolo-gists, engineers, and otologists in devising implantable electronic hearingdevices for the restoration of hearing. The idea perhaps first was conceivedaround 1800 when Alesandro de Volta inserted two metal electrodes, fromthe electrolytic cell he discovered, into his ear canals and reported abubbling sensation. Nevertheless, attempts over the next 50 years to



understand the phenomenon were sporadic. In the mid-1800s, alternatingcurrent was tested, to no avail. Wever and Bray, in 1930, reported that theelectrical response recorded in the vicinity of the auditory nerve of a catwas similar in frequency and amplitude to the sounds to which the earhad been exposed.9,48 One of the first attempts to stimulate the VIIIth nervewas by Lundberg in 1950, who did so during a neurosurgical operation.9,48

Djourno and Eyries, in 1957, placed electrodes on the auditory nerve andstimulated them at different pulse rates, and the patient was able todistinguish such words as papa and allo.9,48 These early observations werethe forerunners of today’s cochlear implant device, the world’s first suc-cessful neural prosthesis. This technology also spun off the developmentof three other types of implantable hearing devices to address the specifictypes of hearing losses prevalent worldwide. Each of the four technologiesseeking to improve the auditory input to the brain over the VIIIth cranialnerve will be reviewed in this chapter. Virtually all types of hearing loss,except for supramedullary brain damage have been ameliorated byimplantable electronic otologic prosthetic devices: bone conductiondevices, middle ear implantable hearing devices (MEIHDs), cochlearimplants (CIs), and auditory brainstem implants (ABIs).

10.2 Review of auditory mechanismsThe ear, of course, provides one of the five senses of the human body. Theouter ear, consisting of the pinna and outer ear canal, is basically the soundgathering portion of anatomy. The input impedance at the entrance to theear canal would be important for the acoustic load of earphones but is oflittle consequence for the consideration of, say, implantable middle eardevices. Our sense of directionality is dependent on differential arrival timesat the outer ear, and the binaural difference in loudness and arrival time atthe auditory cortex of the brain is the neurophysiological basis of determin-ing directionality of sound.

The outer ear is connected to the middle ear by the tympanic membranethrough the ear canal. The middle ear consists of the tympanic membrane,ear ossicles (malleus, incus, and stapes), an attic air space dorsal to the middleear space, the middle ear space containing the ossicles, the antrum with air-containing air cells, and the aditis ad antrum, which connects the antrumand middle ear space. As the impedance-matching portion of the auditorypathway, the middle ear converts variations in air pressure into variationsin fluid pressure (perilymph in the cochlea), where the footplate of the stapesenters the cochlea or inner ear. This impedance mismatch from air to fluidmechanics is accomplished by the middle ear anatomy and mechanics, withthe primary basis of middle ear impedance matching being the surface arearatio of the tympanic membrane and the footplate of the stapes. For normalsound transmission (but not excessively high frequencies) the ossicles act asa unit, a solid body, and the footplate moves as a piston. Hence, the middleear transfer function would involve the acoustic input pressure at the tym-



panic membrane and the middle ear output being the same as the inputpressure in the scala vestibuli of the cochlea (through the stapes footplate).Transfer function and impedance characteristics are important for middleear hearing devices when the ossicular chain is preserved, residual hearingis unimpaired and the ability of high fidelity of amplification is provided bya MEIHD.

The inner ear is the cochlea, a fluid-filled chamber where displacementof the fluids by a traveling wave is converted into a nerve action potentialby hair cells. The 30,000 hair cells act as miniature displacement transducersthat respond to deformations in the perilymph (cochlear fluid). They elicitaction potentials that are tonotopically arranged (in an ordered frequencyarray) both in the cochlea and the brain. Cochlear outer hair cells respondto the traveling wave in the cochlea, have a contractile function (actin fila-ments), and serve as controllable amplifiers for the inner hair cells, whichsend action potentials to the auditory cortex. Loss of outer hair cells resultsin about a 60-dB loss of hearing. The inner ear normally encodes frequenciesby responding to the cochlear fluid movements where different frequenciescause maximum vibration amplitude at different points along the basilarmembrane in the cochlea. The threshold for hearing is 10–11 M of movementby the tympanic membrane with a dynamic range of 120 dB (sound pressurelevel, or SPL). Higher frequencies cause traveling waves in the basal portionsof the basilar membrane, while lower frequencies affect apical regions of themembrane. This is called tonotopic organization (frequency selectivity) andthe cochlear acts as a spectrum analyzer.

In reality, the “sensorineural” hearing loss that is treated by hearing aidsand MEIHDs is the same pathology as is indicated for cochlear implants:loss of inner ear hair cells. In moderate to severe hearing loss, the reductionin hair cell numbers is such that sensitivity to sounds has been reduced;amplified sounds or ossicular movements recruit a greater portion of theremaining hair cells and socially adequate hearing is restored. In profoundhearing loss, the hair cell population is so low that neither acoustic ampli-fication by hearing aids nor mechanical overdrive by MEIHDs can provideadequate restoration of hearing. Direct stimulation of spiral ganglion cellsby passing electrical current across these nerves effectively bypasses the haircell transducers and initiates action potentials to the first relay in the auditorypathway to the brain, the cochlear nucleus. In yet a third form of neuralhearing loss, the VIIIth cranial nerve, the statoacoustic nerve, has been dam-aged or diseased and the auditory pathway interrupted. This most oftenoccurs during surgical removal of acoustic neuromas, tumors that invadethe VIIIth nerve. In this instance, for patients with no VIIIth nerve, neuralprostheses have been designed for provision of an auditory signal by directlystimulating the cochlear nucleus in the brain stem by electrical means. Theauditory brainstem implant is the most recent of neural prostheses to beimplanted for a form of neural deafness. Still, it may be appropriate to referto all of the above forms of deafness simply as “sensory” impairments, ashearing is one of the five senses.



10.3 Hearing lossesApproximately 10% of any nation’s population has a hearing loss of onetype or another. It is estimated that 21.9 million U.S. adults have hearingloss, 15.2 million of those having moderate to severe loss. About 4.7 millionuse hearing aids and 2.3 million are unsatisfied with their hearing aids. Thereare two basic mechanisms for hearing loss. In one, conductive hearing loss, themechanically conductive pathway to the inner ear is impaired, although theinner ear works fine. The second loss mechanism, sensorineural hearing loss,involves the nervous system, meaning that the conduction to the inner earis operative but the transduction mechanism to a nerve action potential byhair cells is impaired; the sound signal does not get conveyed to the auditorycortex of the brain. When devices are implanted to amplify or restore hearing,they deal with mechanics or neural stimulation, depending upon the hearingloss type. In some individuals, there is a mixed hearing loss, meaning bothconductive and neural impairments exist.

The etiology of hearing loss is varied. Chronic infections of the middleear can lead to erosion of the ossicles (conductive loss) or bacterial or viralinvasion of the inner ear, causing loss of hair cells (sensorineural loss). Someantibiotics containing aminoglycosides (used to treat infections) are ototoxicand kill hair cells. Today, five genes are known to be related to hearing, andgenetic causes of hearing loss are just being understood. For example, ahereditary defect on the gene connexin-26 creates a protein necessary for theconveyance pathway of potassium within the inner ear. Potassium contrib-utes to the large generator potential of the hair cells and the absence of thatgene results in hearing loss. Hair cell damage due to excessive noise (noisepollution) is a more recent phenomenon in our younger populations becauseof portable audio systems and high-fidelity, powerful, stereophonic ampli-fiers. Other chemical pollutants or oxygen radicals from air, water, foods, orsmoking are suspect in hearing loss, as they may adversely affect hair cells.Outright damage to the ear from trauma is another cause of hearing loss.Finally, aging is correlated with hearing loss, presbycusis possibly beingrelated to long-term exposure to any or all of the above or simply the normalloss of hair cells with age.

Hence, approaches to restoring hearing by implanted devices and directintervention into the sensory nervous system require an understanding ofthe mechanism of loss and whether the implantable hearing device amplifiessound to a reduced population of hair cells, restores the mechanical connec-tion to a normal population of hair cells, or restores the ability to initiateaction potentials to the brain in an otherwise normal cochlea (auditory nerveis present), except for the absence of hair cells.

10.4 Types of implantable hearing devicesThe four categories of implantable hearing devices (IHD) are classifiedaccording to the types of hearing losses in individuals. These categories



are bone conductors, middle ear implants, cochlear implants, and auditory brain-stem implants. When conductive hearing loss is present, a bone conductionIHD such as the bone-anchored hearing apparatus (BAHA

; Entific Med-ical Systems, Gothenberg, Sweden) seeks to restore hearing by vibratoryconduction through the skull to the inner ear. Bone conduction devices areintended for use with conductive hearing loss, whereas the remainder ofthe IHDs seek to restore hearing due to sensorineural hearing loss. Sensoryhearing impairment is addressed by MEIHDs and the cochlear implant.MEIHDs treat moderate, moderate-to-severe, and severe hearing losses byamplification of sound signals through implanted transducers. Theseamplification mechanisms utilize either piezoelectric crystals or electro-magnetic transducers where greater displacement of the footplate of thestapes will activate a greater number of the (reduced) hair cell populationand thus amplify sounds. Cochlear implants treat profound sensory hear-ing loss and such patients essentially hear nothing without their implantbeing activated. The hair cell population is essentially decimated withprofound hearing loss and only the fibers of the spiral ganglion cells thatproject to the auditory nerve remain. The CI provides direct bipolar ormonopolar electrical stimulation to the spiral ganglion cells according totheir tonotopic (frequency) arrangement within the cochlea. Finally, totaldeafness due to the loss of the VIIIth nerve during tumor removal is treatedby direct electrical stimulation of the brainstem cochlear nucleus, the firstrelay of the VIIIth nerve en route to the auditory cortex. The auditorybrainstem implant (available from Cochlear Corp.; Englewood, CO) pro-vides multielectrode stimulation of the surface of the cochlear nucleus.

10.4.1 Bone conduction devices

Currently, one implantable bone conduction device is commercially avail-able, the bone-anchored hearing apparatus (BAHA

) (Figure 10.1). It hasbeen widely used in Europe since 1977 and is FDA-approved in the UnitedStates.31,49,50 This percutaneous titanium system is implanted in more than7000 patients as a hearing prosthesis and for the attachment of prostheticnoses, ears, eyes, etc. The attachment is based on the physiological phenom-enon of osseointegration (or osteofixation), where the titanium interfacebecomes attached to living bone tissue through the cell matrix ground sub-stances such as glucose aminoglycans. The principle of bone conductionamplification is that an external sound-processing device converts auditorysignals into mechanical vibratory signals and conveys them to the percuta-neous post that is osseointegrated into the post-auricular region of the tem-poral bone. Basically, the output speaker of a hearing aid circuit is physicallycoupled to the titanium post. There is no airborne sound, but the vibrationsin the skull are conveyed to the inner ear and hair cells are bent by thenormal physiological mechanism. The middle ear is bypassed. Patients withconductive hearing loss up to an average bone threshold of 70 dB or whoare unable to wear an air-conduction hearing aid can benefit from the BAHA.



It was recently approved by the FDA for bilateral implantation and attemptsare being made to amplify sounds using more powerful systems (BAHACordelle II) so as to partially restore both conductive and sensorineuralhearing losses.46

10.4.2 Middle ear devices

Ten middle ear implantable hearing devices (MEIHDs) are in active stagesof research, development or commercialization at this time globally: Uni-versity of Dundee, in Dundee, Scotland; IMPLEX

GmbH Hearing Tech-nology, in Ismaning/Munich, Germany; Otologics LLC, in Boulder, CO;Rion Corporation, in Tokyo, Japan; St. Croix Medical, in Minneapolis, MN;University of Sheffield, in England; SOUNDTEC, Inc., in Oklahoma City,OK; Symphonix Devices, Inc., in San Jose, CA; Technical University, inDresden, Germany; and University of Virginia, in Charlottesville, VA. Fivedevices use electromagnetic or motor actuators19 for mechanical micro-movements of the ossicular chain and five devices use piezoelectric ceramicor piezoelectric–hydroacoustic actuators.47 Electromagnetic devices arecharacterized by high efficiency and variable placement of the implantmagnets. Piezoelectric MEIHDs are characteristically small and simple,electronically speaking, but are also less efficient than their magnetic coun-terparts. An advantage is that piezoelectric devices have been made totallyimplantable, as early as the 1970s.

Middle ear implants are one of the newer technologies in hearingrestoration but the concepts behind the technology are the oldest, the firstelectromagnetic hearing amplification having been tested in 1935. The first

Figure 10.1 Bone-anchored hearing aid (Entific Medical Systems; Gothenberg, Swe-den). The external sound processor and percutaneous abutment and titanium implantscrew are shown on the left; the abutment screws into an osseointegrated fixture inthe temporal bone behind the ear. The diagram on the right shows the sound transferpathway to the inner ear through the temporal bone. Vibration of the cochlear capsuleactivates the hair cells.



clinically wearable devices were made in the early 1970s in Japan and late1990s in Germany. The first FDA-approved device was implanted in 200044

followed by a second in 2001,24 with two other devices late into clinicaltrials at this time. The current MEIHD application is for sensorineuralhearing loss in ears with normal middle ear mechanics, which is charac-teristic of the greatest number of hearing-impaired people worldwide.32

Functional gain is the most important rehabilitation parameter for allIHDs. For MEIHDs, the position of the FDA is that comparisons for efficacyand safety should be made with the nominal or best alternative hearingtechnology available to a hearing-impaired patient, in this case a hearingaid. That is, the gain above that of the (control) optimally fit hearing aid (orother device) is the functional gain criterion for device evaluation of efficacy.Today’s high-performance, digital, programmable hearing aids provide arigorous standard of comparison for substantial equivalency. MEIHD com-parisons against either unilateral or bilateral hearing aids for equivalency inperformance are still a matter of investigation. Additionally, a new standardis emerging for comparison of different IHD performances: laser Dopplerinterferometry. This contactless means of measuring displacements is beingused on human temporal bones for middle ear mechanical evaluations andfunctional gain ex vivo.16

IMPLEX GmbH, Munich, Germany, has been commercializing the totallyintegrated cochlear amplifier (TICA

) (Figure 10.2).29,54 Having received theCE mark with its first implantation in Europe (June 1998), the TICA consistsof a totally implantable system with a microphone, rechargeable lithiumbattery, sound processor, and piezoelectric driver with a titanium-couplingrod that vibrates the head of the stapes or incus. Because of the necessaryosseointegration of the coupling rod to produce the push–pull mechanicaltransduction that sets the fluid in the cochlea in motion, the TICA design isbeing modified for a hook-type attachment to the incus, stabilized withionomeric cement, for more assurance of mechanical connection betweenactuator and ossicle. The microphone is located under the skin in the ear

Figure 10.2 Totally integrated cochlear amplifier MEIHD (IMPLEX GmbH; Munich,Germany). On the left is shown the implant portion with implantable microphoneand driver. Shown on the right is the implanted device with the microphone abovethe ear canal and piezoelectric driver on the incus.



canal for detection of sounds gathered by the outer ear and is especiallydesigned with a silicone rubber collar to prevent tissue overgrowth onto thediaphragm of the microphone. Some occurrences of acoustic feedback (theeardrum driven by the transducer and becoming a speaker) due to themicrophone pickup in the ear canal have necessitated partial ossicular dis-connection in some patients. The rechargeable battery allows 60 hours ofcontinues use by patients and requires 90 minutes to recharge the battery,which has a projected lifetime in vivo of 5 years. Such a direct-drive MEIHDhas a maximum power output of 90 dB at 4000 Hz and a maximum functionalgain of 50 dB at 3000 Hz. Functional gain is the improvement over theirpreexisting condition or control comparison.

At this time, at ten sites in the United States, Otologics, Inc., is con-ducting Phase II clinical trials on the middle ear transducer ossicular sim-ulator (MET

), currently a partially implantable device (Figure 10.3).14 Animplantable transducer designed around an electromagnetic motor withassociated electronics and an externally attached Button

(external audioprocessor) drive the ossicular chain by a thin probe that is surgically placedonto the incus in a laser-drilled hole. The aluminum-oxide vibrating probeis connected to the incus by a fibrous union (scar tissue). The Button isattached to the patient’s head by transcutaneous magnetic coupling. Likethe TICA, the MET contains a digital sound processor (two channels and12 digital filter bands) that is programmable by means of NOAH-basedfitting software.

Rion Co., Ltd., and Sanyo Electric Co., Ltd., in Japan collaborated withthe Japanese government and pioneering otologists Suzuki and Yanagiharabeginning in 1978.20,28,47 In 1983, they began clinical investigations, producingthe first clinically wearable MEIHD (Figure 10.4). Using piezoelectric tech-nology, they developed two types of implantable hearing aids (IHAs), apartially implantable hearing aid (PIHA) and a totally implantable hearing

Figure 10.3 Middle ear transducer ossicular stimulator MEIHD (Otologics, Inc.; Den-ver, CO). The implant portion (left) contains a receiving antenna, demodulatingcircuitry, and electromagnetic driver on the end of the cable. The external Buttonsound processor (right) contains microphone, battery, transmitting antenna, and mag-net for transcutaneous attachment.



aid (TIHA). Two types of implantable batteries were developed, a recharge-able and non-rechargeable type. Noteworthy was the implantable recharge-able battery, which became a forerunner of IHD technology. The IHAs weredesigned for patients with chronic middle ear dysfunction and, interestingly,all the recent MEIHD designs are made keeping in mind those patients forwhom middle ear function is normal. A piezoelectric bimorph is attachedto the head of the stapes and driven by an amplifier. In the PIHA, an externalbehind-the-ear receiver contains microphone, battery, sound processor, andtransmitting antenna. The received environmental sounds are processed andtransmitted transcutaneously by radio frequency to the receiver demodula-tor unit under the skin. This internal receiver drives the piezoelectricbimorph as an ossicular vibrator. In Japan, approximately 100 patients havebeen implanted with either the TIHA or PIHA, and U.S. FDA approval isbeing sought.

St. Croix Medical recently completed Phase I clinical trials and wasapproved in 2001 by the FDA to begin Phase II studies on its Envoy

system,a totally implantable MEIHD (Figure 10.5). The Envoy uses piezoelectricceramic transducers placed on the malleus or incus and stapes. A first piezo-electric transducer (the sensor) on the malleus or incus detects tympanicmembrane motion in response to sound. The sound signal is processed bya programmable digital circuit and is used to drive the second transducer(the driver) that is attached to the stapes by ionomeric cement. Programming

Figure 10.4 Partially implantable MEIHD (Rion Co.; Tokyo, Japan). The implant por-tion is shown on the left with a mounting fixture to aid placement of the piezoelectricdriver on to the head of the stapes. The external sound processor (right) contains amicrophone, power supply, and radiofrequency transmitting circuitry for conveyingthe sound signal to the implantable portion.



of the implanted digital chip by an audiologist is accomplished with bidi-rectional telemetry. Personal programming also utilizes the telemetry link toallow the patient to change volume, select a desired environmental setting,or place the Envoy in standby mode. Normally, the incus is interposedbetween the malleus and stapes, but, to prevent mechanical feedback to thesensor, the incus is surgically resected, interrupting the normal conductivepathway to the cochlea. Analogous to acoustic feedback in hearing aids, theoutput transducer could drive the input sensor and create oscillatory feed-back unless the pathway was interrupted.

The Direct System

is manufactured by SOUNDTEC, Inc., and is anelectromagnetic, partially implantable system, the only implant componentbeing a rare earth, permanent magnet in a titanium canister and attachedby fibrous tissue around the incudo–stapedial joint (Figure 10.6).16,24 Theexternal sound processor consists of analog or digital sound processor,battery, microphone, and electromagnetic coil, packaged for behind-the-hear or in-the-canal wear by the patient. The external electromagnetic coil,currently worn in the ear canal, produces a magnetic field that interactswith the permanent magnet implant across the intact tympanic membrane.The Direct System, approved by the FDA for implantation in 2001, is theonly contactless MEIHD with an intact ossicular chain, and the principle

Figure 10.5 Envoy system MEIHD (St. Croix Medical; Minneapolis, MN). Totallyimplantable, the sound processor is powered by a pacemaker-type battery. The sensortransducer is a piezoelectric pickup for vibrations from the malleus or incus. Soundsare processed and conveyed to the piezoelectric driver transducer (also piezoelectric),attached by probe to the stapes.



of operation is like that of an acoustic speaker, with a driving coil andmagnetic driver directly driving the ossicular chain mechanically. Middleear mechanics demonstrate that the closer a transducer is located to thefootplate of the stapes, the higher the frequency response. This is due tothe rocking motion of the stapes at higher frequencies. The magneticimplant of the Direct System is the closest to the stapes footplate of allMEIHDs, with an intact ossicular chain producing the high-frequencyamplification sought after with this technology. Phase II clinical trials atten U.S. sites were completed on 103 patients, and Canadian regulatoryagency approval is also anticipated late 2001.

The Vibrant Soundbridge

, manufactured by Symphonix Devices, Inc.,was the first MEIHD to attain FDA approval for marketing in the UnitedStates (2000) and Europe (CE mark) (Figure 10.7).43,44 Approximately 667patients worldwide, 110 in the United States, have received the VibrantSoundbridge, which has a unique electromagnetic configuration. The float-ing mass transducer is an integrated magnet-coil unit that produces theauditory frequency vibrations on the incus where it is attached. Semi-implantable, the Vibrant Soundbridge system has an implantable electronicspackage that receives both power and the auditory signal through a trans-cutaneous radiofrequency link. The signal transmitted across the skin is viaan amplitude-modulated signal to the internal receiver in the vibrating ossic-ular prosthesis (VORP; the electronics package). The VORP contains a mag-net for transcutaneous attachment to the skin and antennae alignment, ademodulator, and a cable connecting to the FMT. The auditory processorcontains a transmitting coil and microphone, with programmable multiple-band digital signal processor, modulator circuit, and battery.

Figure 10.6 Direct System MEIHD (SOUNDTEC, Inc.; Oklahoma City, OK). Thissemi-implantable electromagnetic device is shown on the left; the sound processorand electromagnetic coil are worn in the ear canal. The magnet implant portion issurgically placed through the ear canal and attached around the incudostapedialjoint. The implant portion (right) is a permanent magnet hermetically sealed in atitanium canister.



Technical University in Desden has been exploring a hydroacoustictransmission principle where vibrations from a piezoelectric driver are con-veyed to a fluid-filled conductor that is in contact with the perilymphaticfluid of the cochlea.26 The ossicular chain is intact and, presumably, becausethe air-to-fluid impedance matching function of the middle ear is bypassedand compensated for by this hydroacoustic transducer, the MEIHD demon-strates a high-frequency response.

Using guinea pigs and human temporal bones, Pellegrin Hospital atUniversity of Bordeaux12 has been investigating a piezoelectric vibratorapplication where the vibrator is placed onto a round window and theossicular chain is left intact. A platinum ball tethered on the end of thepiezoelectric bender bimorph is (slightly) spring loaded onto the roundwindow; thus, no attachment device or polymer is necessary.

The Case Western Reserve University32,33 researched a semi-implantable,middle ear, electromagnetic hearing device (SIMEHD) during 2 years of PhaseI clinical testing. The functional gain observed was not superior to the controlacoustic hearing aid and the project was discontinued. In the SIMEHD, animplanted electromagnetic coil interacted with a magnet attached to the incususing dental cement. In more recent investigations, the magnet, coupled to thetympanic membrane, is being used as a mechanical sound detector. It interactswith the electromagnetic coil, the changing magnetic environment of the coilbecoming a sensor. This detected signal can then be processed, converted intoelectrical stimuli, and presented to intracochlear electrodes for cochlear implantstimulation. Thus, their original MEIHD technology is now being investigatedfor inclusion in a totally implantable cochlear implant device.

The University of Virginia, Charlottesville,45 has also been investigatinga middle ear electromagnetic device; in this instance, a magnet is placed on

Figure 10.7 Vibrant Soundbridge MEIHD (Symphonix Devices, Inc.; San Jose, CA).This semi-implantable electromagnetic device is shown on the left, where a trans-ducer (magnet and encircling coil) is attached to the incus. The internal receiverdemodulates an radiofrequency signal containing frequency and amplitude informa-tion. The middle ear placement is depicted on the right; the hardwired, hermeticallysealed transducer is attached to the incus.



the round window membrane of the cochlea. This membrane moves in theopposite direction of the oval window membrane under the footplate of thestapes. Thus, it too is a reflection of the cochlear fluid movement that activateshair cells and produces the transduction from fluid movement into afferentaction potentials toward the cochlear nucleus. Mechanically sound in princi-ple, the biomedical hurdle remains one of permanent attachment of the mag-net to a constantly growing epithelium of the round window membrane.

The University of Dundee1 MEIHD research is directed toward thedesign of hearing implants and prostheses for ear surgery using a multilay-ered piezoelectric actuator for directly driving the ossicles. This work hasbeen carried out in collaboration with the University of Edinburgh, formingthe collaborative United Kingdom Middle Ear Research Group.

Xomed-Treace, Inc.,23 was the first commercial venture in the UnitedStates to design a MEIHD, and the company received the first FDA investi-gative device exemption to clinically test the electromagnetic device in theearly 1980s. In collaboration with the Hough Ear Institute (Oklahoma City,OK), this venture has employed the magnetic coupling technology that isnow in use worldwide for positioning cochlear implants, as well as MEIHDsand auditory brainstem implants. This approach first used rare earth magnetformulations placed onto the ossicular chain. Early failures with hermeticityand excessive mass loading on the ossicular chain led to corporate abandon-ment of the project by Xomed-Treace. This beginning, however, became theforerunner of the SOUNDTEC Direct System now entering commercializa-tion almost two decades later. The pioneer use by Xomed-Treace and HoughEar Institute of reduced-mass and increased-strength rare earth magnets forsound transduction at the incudo–stapedial joint and their determination ofossicular mass loading limits became critical constituents in the success oftoday’s MEIHD technology.

In the 1980s, Gyrus-ENT22,27,35,51 worked on a partially implantable elec-tromagnetic design that was directed toward patients with chronic middle eardisease and, as such, had partial destruction of the ossicular chain. Becausethere was a conductive loss as well, their design incorporated a permanentmagnet (samarium cobalt) into either a partial ossicular replacement prosthesis(PORP) or total ossicular replacement prosthesis (TORP). Hence, the ossicularchain was reconstructed with a prosthesis that was magnetically responsiveto an electromagnetic field emanating from the ear canal and crossing thetympanic membrane. Difficulties encountered during the limited clinical trialsfor this device included alignment of the magnet in the prosthesis with theelectromagnetic coil and electromagnetic coupling, sizing of the implant toallow for vibratory movements, and preloading of the ossicular replacement.A corporate decision was reached to discontinue the research.

NanoBioMagnetics, Inc., in collaboration with the University of Okla-homa Health Sciences Center, is exploring applications of nanotechnologyin MEIHDs. Additional miniaturization of implantable device componentscan lead to increased safety, performance, and implant life. No results havebeen published to date.



Stanford University19 was the site for some of the pioneering electro-magnetic MEIHD concepts. Investigators glued an AlNiCo magnet to theumbo of the malleus on the outside of the tympanic membrane and placedan electromagnetic coil either in the ear canal or outside of the ear on thepost-auricular skin. This same basic concept of driving a movable permanentmagnet with a fixed electromagnet was tested acutely on patients undergo-ing other procedures in the operating room where the magnet was placedon the incus and round window membrane. No mechanical advantage overthe umbo site was observed. However, when the magnet implant wasclipped to the malleus, one subject wore an 85-mg implant for 22 monthswithout adverse effects or changes in air conduction hearing thresholds upto 400 Hz. This work led to important discussions regarding mass loadingof the ossicular chain. Early data were related to the limitations of loadingthat would impair residual hearing of a person with an intact ossicular chainand a MEIHD.

Resound Corporation41 investigated the placement of a disc-shapedmagnet onto the outside of the tympanic membrane. This magnet was thendriven using the same principle used in research at Stanford Universityand Case Western Reserve University and for the SOUNDTEC Direct Sys-tem. For the Ear Lens, as it was called, an electromagnetic coil was placedexternally at the ear level or around the neck as a collar. This non-implantedIHD held the magnet onto the tympanic membrane by surface tension ofa drop of oil between a custom-molded silicone rubber “lens” and thetympanic membrane. Shortcomings in the available battery power neces-sary to drive the large electromagnetic coils caused this commercial ventureto be discontinued.

Sheffield University2 is investigating the direct drive approach usingan integrated electromechanical device where the electromagnetic coil anda central magnet (like an acoustic speaker diaphragm with central magnetand surrounding coil) are in direct contact with the head of the stapes. Inthis MEIHD, the coil is linked to an external amplifier and microphoneeither by direct connection outside of the ear or by a radiofrequency trans-mission link. The required driving forces for mechanical stapes vibrationare in the range of 0.16 to 16

µN and must produce amplitudes of the stapesbetween 0.1 and 10 nm.

Based on auditory physiology, MEIHD technology has several limitations:

1. One cannot regain frequency sensitivity by pushing the basilar mem-brane harder.

2. When the cochlear amplifier (outer hair cells) is lost, internal noiseis generated by external amplification.

3. Prolonged overdrive of the diminished population of hair cells mayexacerbate the underlying hearing loss.

For long-term restoration of hearing, then, cochlear implant technology maybe the optimal methodology.




The technology underlying the 30 or more cochlear implant devices inves-tigated over the last 20 years sought to stimulate the available neurons ofthe auditory nerve (spiral ganglion cells) by passing current in the vicinityof those neurons.30,39 The challenge included not only safely electrically depo-larizing auditory neurons but also conveying meaningful information aboutspeech to the auditory cortex. Preclinical research on cochlear physiology,electrode design, current delivery, stimulus paradigms, speech encoding,and biomedical engineering has taken place in the United States,25,37,42,53

Australia,8 France,7 Austria, United Kingdom,52 Switzerland,10 Denmark,West Germany,3 and Spain.4

The two physiological hypotheses for auditory frequency discriminationhave been modeled in CI technology. The place principle states that the basilarmembrane of the cochlea has the ability to separate out different frequenciesfrom complex sounds. It states that neurons have characteristic frequenciesand that the VIIIth nerve has a tonotopic organization. The temporal principlestates that a time pattern in sounds can convey information about the soundspectrum (frequency distribution of energy) within that sound to the audi-tory cortex. How the spectra of sounds are distributed in the auditory cortexis the basis of speech discrimination. Spectral analysis by the normal cochleais also dependent on the intensity of sounds. Both principles are involvedin hearing and were considered in the design of speech processing algo-rithms for the CI. For example, localized, controlled, charge-balanced, bipha-sic waveforms are used to depolarize selected pools of neurons accordingto their frequency specificity along the cochlea. Stimulating different elec-trodes depending upon the frequency of the signal restores the filteringfunction of the cochlea and presents electrical stimulation coded for inter-pretation of loudness and pitch information.

Common features of a contemporary CI device include a microphone,external sound processor and power supply, transmitting circuitry,receiver/stimulator package, and electrode array. The microphone picks upsounds and the sound processor filters and selects information that is con-verted into electrical signals that are transmitted to the intracochlear array ofelectrodes. For example, the voicing frequency may be coded as the electrodepulse rate.18 Both the encoded signal and power are transmitted transcutane-ously, using radiofrequency antennae, to a demodulator that assigns thespeech information to the electrode array. Single-channel systems use onlyone electrode, while multichannel systems employ up to 31 electrodes. Vari-ations occur in the packaging and location of external sound processors,microphones, and power supplies, but an attracting pair of magnets on thesurface of and under the skin are usually used to align the antennae and holdthe external devices onto the head. Cochlear implant devices differ in elec-trode design, type of stimulation, and mode of signal processing. Contempo-rary CI systems include AllHear, Inc., in Aurora, OR; Clarion

, AdvancedBionics, Inc., in Sylmar, CA (Figure 10.8); Nucleus

, Cochlear Corporation, in



Englewood, CO (Figure 10.9); Digisonic

, MXM Co., in Vallauris, France;Ineraid, Symbion, Inc., in Provo, UT; Laura Flex, Antwerp Bionic Systems, inBelgium; and COMBI

-40+, MED-EL Corp., in Insbruck, Austria (Figure10.10). In the research phase is the University College London implantabledevice (UCLID), employing a percutaneous connector link between thespeech processor and electrode array. The Clarion, Nucleus and COMBI-40+are approved for patient use in the United States.

Figure 10.8 Clarion II Bionic Ear system CI (Advanced Bionics, Inc.; Sylmar, CA).The ear-level sound processor (left) contains microphone, battery, sound processingcircuitry, transmitting antenna, and magnet for attachment to the head of a patient.The implant (right) contains a receiving antenna, demodulating circuit, and planarcontact electrode array (31) at the end of the cable. Both Clarion and Nucleus electrodearrays are curved to fit inside the cochlear canal.

Figure 10.9 Nucleus 24 Contour CI (Cochlear Corp.; Englewood, CO). The ear-levelsound processor (right) contains a microphone, battery, sound processing circuitry,and transmitting antenna with magnet in its center for attachment to the head of apatient. The implant (left) contains a receiving antenna, demodulating circuit, ground(ball) electrode, and semibanded electrode array (22) at the end of the second cable.



10.5 Signal processingSignal processing strategies, the conversion of the speech signals into elec-trical stimuli, vary from device to device. Processing algorithms over 30 yearssought either to preserve waveforms or preserve speech envelope informa-tion or preserve spectral features from speech (such as formants). Develop-ment of single-channel CI devices began in the 1970s (House/3M and itslater iteration, Vienna/3M single-channel implants25); they are relatively sim-ple in design and cost, using a single electrode, but they became a clini-cal–engineering controversy based on the principle of channel theory andadditional information transfer that became possible through multichannelCI systems. Loudness and temporal variations in speech but limited spectralinformation were conveyed to deaf patients but their open set speech rec-ognition scores were extremely limited. Important information is containedin the speech signal up to 4000 Hz.

10.5.1 Cochlear electrode types

Electrode design for the CI critically evaluates their placement, number andspacing, orientation with respect to spiral gangion cells, and configuration.53

Usually placed within the scala tympani (intracochlear), electrode stimulusparadigms preserve the “place” mechanism of the cochlear for coding fre-quencies. The multichannel electrode array of a CI consists of a flexiblesilicone rubber carrier, in some instances, shaped to fit inside of the scalatympani, with noble metal (usually platinum) electrodes spaced within thebony modiolus of the cochlea. Because high-frequency neurons are at thebasal end of the cochlea, the further the electrode insertion, the lower thefrequency response of the stimulation. Most electrode arrays do not extendmore than 30 mm into the scala tympani so as to minimize damage tocochlear neurons from insertion trauma. The number of electrodes and spac-

Figure 10.10 COMBI-40+ CI (MED-EL Corp.; Insbruck, Austria). The ear-level soundprocessor (upper) contains a microphone, battery, sound processing circuitry withear hook for attachment to the ear, and transmitting antenna with magnet in its centerfor attachment to head of a patient. The implant (below) contains a receiving antenna,demodulating circuit, ground (ball) electrode, and split electrode array (12 pairs) atthe end of the second cable.



ing between them affects the resolution for coding those frequencies, withinthe constraints of the number of surviving neurons and the spread of elec-trical excitation. Electrode design seeks to discretely depolarize a local pop-ulation of neurons.15 Approximately 30,000 branches (normal cochlea) of theauditory nerve theoretically can be divided and assigned frequencies by thereceiver– stimulator–electrode design. In reality, current spreads symmetri-cally in the cochlea, and an isolated set of neurons is difficult to stimulate.This is especially so with monopolar stimulation but constrained to a degreewith bipolar stimulation. Channel theory predicts that separation of neuronaldepolarizations within the cochlea will facilitate frequency analysis andspeech discrimination by the cortex.36

Currently, CI devices offer either monopolar or bipolar electrode araysor both. The AllHear is a single monopolar CI. The Digisonic has 15 chan-nels for intracochlear implantation and a multiarray with three insertionleads for ossified cochlea where spiral ganglion cells are positionedbetween electrode arrays for depolarization. Ineraid, the only percutaneouscochlear implant device, is no longer manufactured; it used six electrodesspaced 4 mm apart, with the four apical electrodes in monopolar configu-ration. The Nucleus device uses up to 22 electrodes spaced 0.75 mm apart;electrodes 1.5 mm apart are used as bipolar pairs. The Clarion, CII BionicEar

also provides both monopolar and bipolar configurations and usesup to 31 electrodes. The COMBI-40+ uses 12 pairs of electrodes equallydispersed over 26.4 mm (C40+S) or a split compressed electrode array offive and seven pairs, to be used in an ossified cochlear where drilling andelectrode insertion are performed in a monopolar configuration. TheUCLID employs 8 to 22 active electrodes. Recent CI developments placethe electrode array as close to the inner wall or spiral ganglion of the VIIIthnerve as possible.

10.5.2 Types of cochlear stimulation

The two types of stimulation, analog and pulsatile, depend on how infor-mation is presented to the electrodes. In analog stimulation, the acousticwaveform is replicated and presented to the electrode. For multichannelsystems, the acoustic waveform is bandpass filtered and presented to all theelectrodes simultaneously in analog form.38 The neural plasticity of the brainis allowed to sort out the information and formulate a meaningful auditorypercept.40 Neurophysiologically, a disadvantage of this simultaneous analogstimulus paradigm is channel interaction, or a lack of discretization of infor-mation. In pulsatile stimulation, electrodes deliver a narrow set of biphasicand charge-balanced pulses, the amplitude being related to the height of theenvelope of the filtered waveforms. Rate of stimulation affects speech rec-ognition performance and this is likely due to the pulses not overlapping(or not being simultaneous). Both analog and pulsatile stimuli can be senttranscutanously by radiofrequency transmission or percutaneously by hard-wire as with the Ineraid.



Multichannel CI devices began to appear in the 1980s and immediatelyquestions arose as to how many electrodes were best and what kind ofinformation should be transmitted to each electrode. The Ineraid first usedthe compressed analog (CA) speech-processing algorithm. In CA, the signalis first compressed with automatic gain control, then filtered into four con-tiguous frequency bands with center frequencies. The filtered waveformshave individual gain controls, then are fed to the four intracochlear elec-trodes. The CA algorithm uses analog stimulation, delivering continuousanalog waveforms to four electrodes simultaneously. Channel separation hasnot been optimally demonstrated with CA processing.

The next iteration of speech processing algorithms in multichannel CIdevices is a continuous interleaved sampling (CIS) strategy, using non-simul-taneous, interleaved pulses.17,55 The pulse amplitudes were derived from thespeech envelopes, which were derived by full-wave rectification and low-pass filtering. Several CIS parameters have been optimized over the yearsthat have increased speech recognition in quiet from the initial single-channelresults of 5 to 15% to over 80% today. Pulse rate and pulse duration at eachelectrode vary from patient to patient so the programmable receiver–stimu-lator is mapped for each patient. Pulses between 100 and 2500 pulses/secare being used at approximately 33

µsec/pulse. Sound delivery (non-simul-taneous) by speech processors can be 18,200 pulses/sec (COMBI-40+), 14,000pulses/sec (Nucleus 24 Contour

), or 250,000 pulses/sec (Clarion CII BionicEar). Stimulation order of the electrodes has been used to optimize channelseparation,5 sometimes from apex to base of the cochlea and sometimesstaggered to maximize spatial separation between stimulated electrodes. Thecompression of the speech envelope outputs is the important determinantof pulse electrical amplitudes. The speech-processing algorithm must ensurethat the range of acoustic envelope amplitudes conforms to the auditorydynamic range of the patient. Dynamic range is the range between barelyaudible speech and uncomfortably loud sounds. Implant patients may havedynamic ranges of only 5 dB, hence CIS strategies use nonlinear compressionfunctions to fit the patient’s comfort. Such logarithmic functions are pro-grammed into the patient’s map in the erasable programmable read-onlymemory of the speech processor.

In addition to continuously improving speech-processing algorithms,additional improvements have occurred in CI technology. For example,reverse telemetry was first incorporated into the multichannel CI by theNucleus Neural Response Telemetry (NRT

). Here, interrogation by thedigital processor of the cochlear environment can report selective electrodeviability in stimulating the spiral ganglion cells. A brief set of current pulsesis delivered to an electrode while a second electrode nearby records theneural response. This response is essentially an intracochlear recording ofwave I of the electrically evoked auditory response (EABR) that allows anaudiologist to optimally assign electrode stimulations. Thresholds of audi-tory nerve stimulation can also be accurately detected and better establishthe dynamic range of a patient’s hearing.



10.5.3 Auditory brainstem implants

If the auditory nerve has been damaged, a modified CI device can accomplishdirect stimulation of the cochlear nucleus when electrodes are placed in acarrier for placement onto the surface of the medulla, over the cochlearnucleus.11,34 Auditory brainstem implants by Cochlear Corp. and Digisonichave been approved for use in humans, and patient data are just beingcompiled at this time (Figure 10.11).

10.6 SummaryThe types of hearing losses experienced today by every population rangefrom mild to severe, most commonly involving loss of inner ear hair cellsand the transduction method for initiating the auditory percept. This sensoryloss is being ameliorated today, not only by classical acoustic hearing aidsbut by implantable, sophisticated electronic devices. Middle ear, direct drivemechanical systems (applicable for 19 million Americans) are an emergingtechnology after electrical neuroprosthetic cochlear implants, which havelimited application (about 1 million in the United States). Nevertheless, over

Figure 10.11 Auditory Brainstem Implant (Cochlear Corp.; Englewood, CO). Thewearable speech processor (upper) has an ear-level microphone and associated trans-mitting antenna, and the magnet for attachment contains the batteries and soundprocessing circuitry. The implant portion (below) is comparable to a cochlear implantwith a ball and ground electrode but with a flat surface array of electrodes at theend of the second cable that is placed on the medullary surface, over the cochlearnucleus. The implant also contains a receiving antenna, attaching magnet, and de-modulating circuit.



40 research endeavors worldwide have resulted in three leading cochlearimplant devices and six promising MEIHDs. The future technological direc-tions of these neuroprosthetic devices include further miniaturization, totalimplantation, and greater lifetimes in vivo. If recombinant or other geneticmeans of restoring (hair cells) hearing do not emerge, then IHDs will con-tinue as a viable bioengineering approach to hearing loss.

References1. Abel, E.W., Wang, Z.G., Mills, R.P., and Liu, Y., Performance and power

consumption of a multi-layer piezoelectric actuator for use in middle earimplants, Proc. 3rd Int. Symp. on Electronic Implants in Otology and ConventionalHearing Aids, May 31–June 2, Birmingham, 2000.

2. Affane, W. and Birch, T.S., A microminiature electromagnetic middle-ear im-plant hearing device, Sensors Actuators A, 46–47, 584–587, 1995.

3. Banfai, P., Karczag, A., Kubik, S., Luers, P., and Surth, W., Extracochlearsixteen channel electrode system, Otolaryngol. Clin. North Am., 19(2), 371–408,1986.

4. Bosch, J., Prades, J., Colomina, R., and Monferre, A., A model multichannelcochlear implant, Ear Hear., 1(4), 226–228, 1980.

5. Carlyon, R.P., Geurts, L., and Wouters, J., Detection of small across-channeltiming differences by cochlear implantees, Hear. Res., 1(1–2), 140–154, 2000.

6. Chasin, M. and Spindel, J., Middle ear implants: a new technology, HearingJ., 54(8), 2001.

7. Chouard, C.H., The surgical rehabilitation of total deafness with the multi-channel cochlear implant: indications and results, Audiology, 19(2), 137–145.

8. Clark, G.M., Kranz, H.G., Minas, H., and Nathar, J.M., Histopathologicalfindings in cochlear implants in cats, J. Laryngol. Otol., 89(5), 495–504, 1975.

9. Clark, G.M., Tong, Y.C., Patrick, J.F., Cochlear Prostheses, Churchhill-Living-stone, Melbourne, 1990.

10. Dillier, N., Spillmann, T., Fisch, U.P., and Leifer, L.J., Encoding and decodingof auditory signals in relation to human speech and its application to humancochler implants, Audiology, 19(2), 146–163, 1980.

11. Di Nardo, W., Fetoni, A., Buldrini, S., and Di Girolamo, S., Auditory brainstemand cochlear implants, functional results obtained after one year of rehabili-tation, Eur. Arch. Otorhinolaryngol., 258, 5–8, 2000.

12. Dumon, T., Zennaro, O., Aran, J.M., and Bébéar, J.P., Piezoelectric middle earimplant preserving the ossicular chain, Otolaryngol. Clin. North Am., 28(1),173–187, 1995.

13. Fredrickson, J.M., Coticchia, J.M., and Khosla, S., Current status in the devel-opment of implantable middle ear hearing aids, Adv. Otolaryngol., 10, 189–203,1996.

14. Fredrickson, J.M., Coticchia, J.M., and Khosla, S., Ongoing investigations intoan implantable electromagnetic to severe sensorineural hearing loss, Otolaryn-gol. Clin. North Am., 28 (1), 107–120, 1995.

15. Friesen, L.M., Shannon, R.V., and Slattery, III, W.H., Effects of electrode loca-tion on speech recognition with the Nucleus-22 cochlear implant, J. Am. Acad.Audiol., 11, 418–428, 2000.



16. Gan, R.Z., Wood, M.W., Ball, G.R., Dietz, T.G., Dormer, K.J., Implantablehearing device performance measured by laser doppler interferometry, EarNose Throat J., 76(5), 297–309, 1997.

17. Geurts, L. and Wouters, J., Coding of the fundamental frequency in continu-ous interleaved sampling processors for cochlear implants, J. Acoustic. Soc.Am., 109(2), 713–726, 2001.

18. Geurts, L. and Wouters, J., Coding of the fundamental frequency in contiuu-ous interleaved sampling processors for cochlear implants, J. Acoust. Soc. Am.,109(2) 713–726, 2001.

19. Goode, R.L., Current status of electromagnetic implantable hearing aids, Oto-laryngol. Clin. North Am., 22, 201–209, 1989.

20. Gyo, K., Saiki, T., and Yanagihara, N., Implantable hearing aid using a piezoelec-tric ossicular vibrator: a speech audiometic study, Audiology, 35, 271–276, 1996.

21. Håkansson, B., Lidén, G., Tjellström, A., Ringdahl, A., Jacobsson, M., Carlsson,P., and Erlandson, B.E., Ten years of experience with the Swedish bone-anchored hearing system, Ann. Otol. Rhinol. Laryngol., 99(10, suppl. 151, pt.2), 1–16, 1990.

22. Heide, J., Tatge, G., Sander, T. et al., Development of a semi-implantablehearing device, in Advances in Audiology, Vol. 4, Middle Ear Implant, ImplantableHearing Aids, Hoke, M., Ed., Karger, Basel, 1988.

23. Hough, J., Vernon, J., Dormer, K., Johnson, B., and Himelick, T., Experienceswith implantable hearing devices and a presentation of a new device, Ann.Otol. Rhinol. Laryngol., 95(1), 60–65, 1986.

24. Hough, J.V., Dyer, R.K., Matthews, P., and Wood, M.W., Early clinical results,SOUNDTEC implantable hearing device Phase II study, Laryngoscope, 111(1),1–8, 2001

25. House, W.F., Cochlear implants, Ann. Otol. Rhonol. Laryngol., 85(3, suppl. 27),1–93, 1976.

26. Huttenbrink, K.-B., Zahnert, Th., Bornitz, M., and Hoffmann, G., Biomechan-ical aspects in implantable microphones and hearing aids and developmentof a concept with a hydro dynamic acoustical transmission, Acta Otorhino-laryngol., (121), 185–189, 2001.

27. Kartush, J.M. and Tos, M., Electromagnetic ossicular augmentation device,Otolaryng. Clin. North Am., 28, 155–172, 1995.

28. Kodera, K., Suzuki, J.I., Nagai, K., and Yabe, T., Sound evaluation of partiallyimplantable piezoelectric middle ear implant: comparative study of frequencyresponses, Ear Nose Throat J., 73(2), 108–111, 1994.

29. Leysieffer, H., Baumann, J.W., Müller, G., and Zenner, H.P., An implantablepiezoelectric hearing aid transducer, HNO, 45, 792–800, 1997.

30. Loizou, P.C., Introduction to cochlear implants, IEEE Signal Process. Mag,101–130, 1998.

31. Lustig, L.R., Arts, A.H., Brackman, D.E. et al., Hearing rehabilitation usingthe BAHA bone-anchored hearing aid: results in 40 patients, Otol. Neurotol.,22(3), 328–334, 2001.

32. Maniglia, A.J. and Proops, D.W., Eds., Implantable electronic otologic devices:state of the art, Otolaryngol. Clin. North Am., 34(2), 1–522, 2001.

33. Maniglia, A.J., Ko, W.H., Rosenbaum, M., Falk, T., Zhu, W.L., Frenz, N.W.,Werning, J., Masin, J., Stein, A., and Sabri, A., Contactless semi-implantableelectromagnetic middle ear device for the treatment of sensorineural hearingloss, Ear Nose Throat J., 73, 78–90, 1994.



34. Marangos, N., Stecker, M., Sollman, W.P., and Laszig, R., Stimulation of thecochlear nucleus with multichannel auditory brainstem implants and long-term results: Freiburg patients, J. Laryngol. Otol., (27, suppl.), 27–31, 2000.

35. McGee, T.M., Kartush, J.M., Haide, J.C. et al., Electromagnetic semi-implant-able hearing device: Phase I clinical trials, Larynogoscope, 101, 355–360, 1991.

36. Mehr, M.A., Turner, C.W., and Parkinson, A., Channel weights for speechrecognition in cochlear implant users, J. Acoust. Soc. Am., 109(1), 359–366, 2001.

37. Michaelson, R.P., Merzenich, M.M., Schindler, R.A., and Schindler, D.N.,Present status and future developments of the cochlear prosthesis, Ann. Otol.Rhinol. Laryngol., 84(4, pt. 1), 494–498, 1975.

38. Morse, R.P. and Roper, P., Enhanced coding in a cochlear-implant model usingadditive noise: aperiodic stochastic resonance with tuning, Phys. Rev., 61(5),61–66, 2000.

39. Niparko, J.K., Kirk, K.I., Mellon, N.K., Robbins, A.M., Tucci, D.L., and Wilson,B.S., Eds., Cochlear Implants: Principles & Practices, Lippincott, Williams &Wilkins, Philadelphia, 2000.

40. Nishimura, H., Doi, K., Iwaki, T., Hashiwawa, K., Oku, N. et al., Neuralplasticity detected in short- and long term cochlear implant users using PET,NeuroReport, 11(4), 811–815, 2000.

41. Perkins, R. and Pluvinage, V., The ear lens: a new method of sound transmis-sion to the middle ear-current status [abstract 34], presented at the Int. Symp.on Electronic Implants in Otology, November, Orlando, FL, 1993.

42. Rabinowitz, W.M. and Eddington, D.K., Effects of channel-to-electrode map-pings on speech perception with the Ineraid cochlear implant, Ear Hear., 16(5),450–458, 1995.

43. Snik, A.F.M. and Cremers C.W.R.J., First audiometic results with the vibrantsoundbridge: a semi-implantable hearing device for sensorineural hearingloss, Audiology, 38, 355–338, 1999.

44. Snik, A.F.M. and Cremers C.W.R.J., The effect of the “floating mass transduc-er” in the middle ear on hearing sensitivity, Am. J. Otol., 21, 42–48, 2000.

45. Spindel, J.H., Corwin, J.T., Ruth, R.A. et al., The basis for around windowelectromagnetic implantable hearing aid, in Proc. of the 13th Annual Int. Conf.of IEEE Eng. in Med. Biol. Society, October, Orlando, FL, 1991.

46. Stenfelt, S., Håkansson, B., Jönsson, R., and Granström, G., A bone-anchoredhearing aid for patients with pure sensorineural hearing impairment, Scand.Audiol., 29, 175–185, 2000.

47. Suzuki, J.I., Ed., Middle ear implant: implantable hearing aids, Adv. Audiol.,4, 1–174, 1988.

48. Syms, III, C.A. and House, W.F., Surgical rehabilitation of deafness, Otolaryn-gol. Clin. North Am., 30(5), 777–782, 1997.

49. Tietze, L. and Papsin, B., Utilization of bone-anchored hearing aids in chil-dren, Int. J. Pediatr. Otorhinolaryngol., 58(1), 75–80, 2001.

50. Tjellström, A. and Håkansson, B., The bone-anchored hearing aid, Otolaryngol.Clin. North Am., 28(1), 53–72, 1995.

51. Tos, M., Salomon, G., and Bonding, P., Implantation of electromagnetic ossic-ular replacement device, Ear Nose Throat J., 73(2), 92–103, 1994.

52. Walliker, J.R., Rosen, S., Douek, E.E., Fourcin, A.J., and Moore, B.C., Speechsignal presentation to the totally deaf, J. Biomed. Eng., 5(4), 316–320, 1983.



53. Wilson, B.S., Cochlear implant technology, in Cochlear Implants-Principles &Practices, Nikarko, J.K., Kirk, K.I., Mellon, N.K., Robbins, A.M., Tucci, D.L.,and Wilson, B.S., Eds., Lippincott, William & Wilkins, Philadelphia, 2000, pp.109–128.

54. Zenner, H.P., TICA totally implantable system for treatment of high frequencysensorineural hearing loss, Ear Nose Throat J., 79(10), 770–777, 2000.

55. Ziese, M., Stützel, A., von Specht, H., Begall, K., Freigang, B., Sroka, S., andNopp, P., Speech understanding with the CIS and the N-of-M strategy in theMED-EL COMBI-40+ system, Otorhinolaryngology, 62, 321–329, 2000.



chapter eleven

Visual neuroprostheses

David J. Warren and Richard A. Normann

Contents

11.1 Introduction11.2 Anatomy and physiology of the visual pathway

11.2.1 Retinal anatomy and physiology11.2.2 Organization of the optic nerve and tract11.2.3 Anatomy and physiology of subcortical structures11.2.4 Visual cortex anatomy and physiology11.2.5 Pathologies leading to blindness11.2.6 Concluding remarks

11.3 Elements of a visual neuroprosthesis11.3.1 Video encoder11.3.2 Signal Processing11.3.3 Telemetry and power interface11.3.4 Neural stimulator11.3.5 Neural interface

11.3.5.1 Physical biocompatibility11.3.5.2 Biological biocompatibility

11.4 Subretinal neuroprosthetic approach11.4.1 Optobionics group11.4.2 Southern German group11.4.3 Summary

11.5 Epiretinal neuroprosthetic approach11.5.1 Humayun and de Juan group11.5.2 MIT and Harvard group11.5.3 Northern German group11.5.4 Summary

11.6 Optic nerve neuroprosthetic approach



11.7 Cortically based neuroprosthetic approach11.7.1 Historical overview11.7.2 Electrode arrays11.7.3 Animal experiments11.7.4 Human experimentation


11.1 IntroductionIn the eighteenth century, Luigi Galvani, an Italian physician, observed thatdissimilar metals, attached to a frog’s leg and connected together, can causethe frog’s skeletal muscles to contract. Subsequently, natural philosophers,and later physiologists, have come to appreciate that electricity is one of thekey features associated with the control of the body. Based upon the pio-neering work of Einthoven in studying the role of electricity in the sensoryand motor parts of the body, it has become clear that the contraction of allmuscle is associated with local electrical fields and that, by the discriminateapplication of externally applied electrical fields, one can intervene withdamaged or diseased parts of the nervous system and thereby restore lostfunctions. The first, and most lifesaving, example of this electrical interven-tion is the cardiac pacemaker, which was developed in the middle of the lastcentury and has become a standard therapeutic approach to a variety ofcardiac arrhythmias and pathologies.

The extension of this successful therapy to other sensory and motorsystems has fostered the neuroprosthetic field. This field is in its infancy butis experiencing great interest in both academic and commercial sectors. Withthe widespread availability of VLSI (very-large-scale integration) circuitryand with the emergence of microfabrication technologies, researchers arenow able to build interfaces to the nervous system with greater complexitythan could have been imagined only a few decades ago. These interfaces arebeing built of highly biocompatible materials that contain feature sizes com-parable to the neurons that the devices are intended to interact with. Further,our understanding of the function and dysfunction of the nervous systemhas improved to the point that, in many cases, researchers have suggestedplausible interventional routes whereby neuroprosthetic systems couldexpect to be efficacious. It is clearly the right time for this technology.

About a half century ago, Brindley, at Cambridge University in England,conducted a bold set of experiments wherein he tried to restore a visualsense to a few individuals who had become profoundly blind. These pio-neering experiments did not result in the restoration of useful vision inthese subjects, but they did indicate that the concept could be entertainedseriously. Thus, the visual neuroprosthetic field was born. Because the tech-nological innovations described above had yet to be developed, littleprogress was made in the field up until the last decade. Over this last 15



years, many individuals have begun working in this field, and, while clinicalsystems are still probably a decade away, it is becoming clear that suchsystems will likely find their way into the operating rooms of major hospi-tals around the world.

This chapter provides an overview of some of the progress that hasbeen made in this field. We will describe the anatomy and physiology ofthe visual pathways and human-engineered systems that have beendesigned to interface with neuronal ensembles at the levels of the retina,the optic nerve, and the visual cortex. We describe animal experiments thatsupport the safety of these systems, and preliminary human experimentsthat are beginning to demonstrate the efficacy of the approach. It is stressedthat the human experimentation is still in its infancy, and as a result thefindings are mixed and inconclusive, but encouraging. However, theincreased numbers of experimenters working in the area, the accelerationof technological innovation, and the commercial impetus to develop suc-cessful clinical systems make it clear that visual neuroprosthetic systemswill provide, perhaps, the first interventional systems that could restoreuseful vision to those with profound blindness. Useful vision in this contextwill not be vision as enjoyed by normally sighted individuals. First-gen-eration systems will be pixelized, will contain small numbers of pixels, andwill recreate narrow field views of the visual world in front of the blindsubject. However, it is hoped that such systems will allow the subjectindependent mobility without requiring a guide dog or a family memberor friend, at least in familiar environments, and perhaps even in unfamiliarvisual environments.

11.2 Anatomy and physiology of the visual pathwayDespite almost a half century of experimentation, the best site for implement-ing a visual neuroprosthesis has yet to be resolved. As will be seen in subse-quent sections, numerous sites have been investigated, each with its relativemerits and shortcomings. To provide a better understanding of each neuro-prosthetic approach and to give grounds for comparing the competingapproaches, an awareness of the underlying neural system is necessary. Asdetails of the visual pathway are readily available in any elementary anatomyand physiology text, the discussion here is limited to a short review highlight-ing the potential sites for a neuroprosthesis. In particular, we have chosen tohighlight the neural organization of these sites and how this organizationrelates to a neuroprosthetic application. The majority of the information pro-vided in this section comes from either well-known neuroscience texts1,2 orthe excellent Web site http://www.webvision.med.utah.edu.

The visual pathway, highlighting the points where vision neuroprosthe-ses have been or could be implemented is shown schematically in Figure11.1. Light entering the eye falls upon the retina, located at the back surfaceof the eye. Here, photoreceptor neurons convert the electromagnetic energyof the light into electrochemical signals. This is the first stage of a series of



http://www.webvision.med.utah.edu/

retinal neurons that process features of the visual scene. The axons of thelast of the series, the retinal ganglion cells, are collected together into theoptic nerve (ON). These axonal fibers are reorganized at the optic chiasm (OC)and then project to a number of subcortical structures through the optic tract(OT). The majority of these axons form synapses in the lateral geniculatenucleus (LGN) of the thalamus, where another series of neurons furtherprocess the visual scene. In turn, the axons of the LGN neurons projectthrough the optic radiation (OR) to the cerebral cortex. Here, a hierarchicalscheme of visual processing occurs over much of the posterior region of theoccipital cerebral cortex. The regions subserving this processing are collec-tively called visual cortex (VC).

A number of themes are common to the neural organization at all sitesalong the visual pathway. The first of these is that, for any neuron along thepathway, its receptive field describes the type of the visual stimuli that causesthe neuron to respond. The receptive field characterization of a neurondetails the nature of the visual stimulus (location in visual space, shape, size,intensity, color, etc.) that optimally drives the neuron and how the neuron’sactivity changes for other than the optimal stimulus. For example, a partic-ular photoreceptor neuron may reserve its greatest response for a small bluespot of light at a particular location relative to central vision. Most probably,external electrical stimulation of the same neuron could evoke in a similarpercept. A related and very critical theme for a neuroprosthesis is that themap from visual space to neural space is visuotopic. That is, the neurons atany site along the visual pathway are arranged so that their receptive fieldlocations form an organized and approximately linear map of visual space.This implies that nearby objects in visual space evoke activity in nearby

Figure 11.1 Schematic view of the visual pathway, highlighting points where a visualneuroprosthesis has been or could be investigated. This illustration also shows theconcept of visuotopy with the letters A and B. The visuotopic organization of thepathway results in nearby objects in visual space being represented in nearby neuronsthroughout the visual pathway.



neurons. Hence, the outline of a rectangle presented in visual space willresult in activity in a similarly shaped arrangement of neurons. The arrange-ment of neurons may be differentially stretched in each axis, rotated, andevenly slightly warped but still would appear as the outline of a rectangle.Presumably, external electrical stimulation of the same ensemble of neuronswill result in the perception of the outline of a rectangle. A last theme is thatthe visual pathway represents a massively parallel method of signal process-ing. Associated with this parallelization are a number of distinct processingpathways found in primates, including humans. The principal two path-ways, the M pathway (for magno, or large) and the P pathway (for parvo, orsmall), begin at the retina and are segregated throughout much of the visualpathway. It is thought that the M and P pathways represent two broadfeatures of an object in visual space: where the object is located and whatthe object is, respectively. At this point, it is unclear whether a neuropros-thesis should (or could) preferentially evoke activity in the M pathway, theP pathway, or both pathways in order to provide the best percept. Never-theless, due to the segregation of the two pathways, there could be a pref-erence for stimulating one pathway over the other at various points alongthe visual pathway.

11.2.1 Retinal anatomy and physiology

If one were to observe a normal human retina through an ophthalmoscope,one would see the gross morphological features illustrated in Figure 11.2.On the right side of the figure is the head of the optic nerve, also called theoptic disk. A number of blood vessels originate from approximately the centerof the optic disk and spread to cover much of the inner retina (the side ofthe retina closest to the lens). These vessels are supplied by the central arteryand vein of the retina, which pass through the optic nerve. The outer retina(the side of the retina farthest from the lens and not visible in the figure) isperfused by the blood vessels running behind the retina in the choroid. Theblood vessels on the inner retina and in the choroid form the retinal andchoroidal circulations, respectively. To the left of the optic disk, one observesa small (~1.5-mm diameter) darker region of the retina devoid of major bloodvessels. This region is called the fovea and it subserves the approximately 6degrees of central vision, equivalent to a 10-cm-diameter circular region ata distance of 1 m. The fovea and the closely surrounding region are calledthe macula lutea or, more commonly, the macula. As the macular region isnot readily seen in the figure, a dashed line indicates the border of themacula. This region, having a diameter of approximately 5 mm, subservesthe approximately 20 degrees of the central visual field. This is equivalentto a 35-cm-diameter circular region at a distance of 1 m, or approximatelythe entire screen on a standard 15-inch monitor observed at 1 m. As theoptics of the eye are relatively linear, the magnification factor (millimetersof neural tissue per degree of visual space) is approximately 0.25 throughoutthe retina.



The retina has a laminar neural organization, a highly simplified view ofwhich is presented in Figure 11.3. This organization consists of three nuclearlayers (layers of neuronal cell bodies) separated by two plexiform layers(layers of synaptic connections). An inner and outer limiting membrane aswell as an outer epithelial layer surround the neural retina. As the photore-ceptors are located in the outermost nuclear layer in humans, the descriptionlogically starts with the outermost layer. The retinal pigmented epithelium (RPE)is not properly part of the neural retina but bears discussion due to itsassociation with pathologies of the retina. This layer provides critical nutri-ents to the photoreceptors as well as phagocytosis of the exhausted compo-nents of the photoreceptors that are shed on a daily basis. The outer nuclearlayer (ONL) consists of the photoreceptor neurons that convert light’s elec-tromagnetic energy into electrochemical energy. In humans, there are twomorphological classes of photoreceptor neurons, rods and cones. Cone pho-

Figure 11.2 Normal human retina as viewed through an ophthalmoscope, highlight-ing the regional specialization of the retina. The optic nerve head, also known as theoptic disk, is where the axons of the retinal ganglion cells are collected together intothe optic nerve as these axons course towards their subcortical targets. This regionis devoid of photoreceptors. Additionally, the blood vessel of the retinal circulationoriginates from the optic nerve head. The fovea, the region of darker pigmentation,subserves the central 6 degrees of visual space and is the region of greatest visualacuity. The photoreceptors in this region are primarily cones. The macula lutea,outlined with a dashed line, subserves the central 20 degrees of visual space. Bothrod and cone photoreceptors are found outside the fovea, with rod photoreceptorsbecoming more prevalent the farther one is from the fovea. The two principal sourcesof untreatable blindness, age-related macular degeneration and retinitis pigmentosa,are characterized by loss of photoreceptors in the macular region and outside themacular region, respectively. (Adapted from Webvision. With permission.)



toreceptors are functional in typical daytime viewing conditions and mediatecolor vision. Humans have three subclasses of cone receptors (S, M, and L)that are most sensitive to blue, green, and red colors, respectively. Rod recep-tors become functional only at low light levels and, hence, are associated withnight vision. As all rod receptors are most sensitive to a single color (blue-green) in humans, at night we can only distinguish relative intensity. Bothrod and cone photoreceptors utilize graded voltage potentials to signal achange in stimulus intensity. That is, a change in luminance results in a changein the membrane potential, not an action potential. The densities of rod andcone photoreceptors vary with distance from the fovea (or with retinal eccen-tricity). Within the fovea, the photoreceptors are exclusively cones. For thefirst 10-degree radius about the fovea, covering approximately the macularregion, the density of cone photoreceptors drops off rapidly to approximately1/20th the density at the center of the fovea. An inverse relationship occurswith the density of rod photoreceptors, starting with a nil density at the foveathat grows rapidly for the first 20-degree radius about the fovea. The outerplexiform layer (OPL) separates the ONL and the inner nuclear layer (INL) and

Figure 11.3 Schematic diagram of the organization of the retina, highlighting theprincipal cells of the retina and the nomenclature of the neural retina. Light enteringthe eye passes through the entire neural retina before being transduced into electricalsignals by the rod and cone photoreceptors of the outer neural layer. These signalsare processed in a columnar fashion by the bipolar cells of the inner neural layer andthe retinal ganglion cells of the ganglion cell layer as well as being processed hori-zontally by the horizontal and amacrine cells of the inner nuclear layer. The axonsof the retinal ganglion cells are collected together into the nerve fiber layer, whichlies between the ganglion cells and the inner surface of the retina, or the epiretinalsurface. (Adapted from Webvision. With permission.)



is the region where the photoreceptors of the ONL form synapses with thetwo classes of neurons of the INL (bipolar and horizontal neurons). Likephotoreceptors, both bipolar and horizontal neurons utilize graded voltagepotentials. The third class of neurons of the INL, amacrine cells, fire actionpotentials. The inner plexiform layer (IPL) separates the INL and the ganglioncell layer (GCL) and is the region where the bipolar and amacrine neurons ofINL form synapses with GCL neurons. The neurons of the GCL consist almostexclusively of action potential firing retinal ganglion cells (RGC). The axons ofthe RGC form the nerve fiber layer (NFL) as they cross the inner surface of theretina, heading towards the optic disk. These fibers are unmyelinated untilthey reach the optic disk. The inner limiting membrane (ILM) separates theretina from the vitreous humor.

Visual information is processed from the outer to inner layers in a colum-nar functional organization. That is, the activity of a ganglion cell is modu-lated by the photoreceptor and bipolar neurons along a column from theinner to outer retina. This way the visuotopic map is preserved across thelayers of the retina. Additionally, the horizontal and amacrine neurons inte-grate information across the neighboring regions of visual space. The degreeof convergence of information across visual space depends on the retinaleccentricity of the ganglion cell. At the fovea, there is a one-to-two relation-ship between photoreceptors and ganglion cells; hence, the fovea is theregion of greatest visual acuity. As one moves to the periphery, a largenumber of photoreceptors drive a single ganglion cell.

The degree of convergence is reflected in the receptive field propertiesof the RGC. At the fovea, the receptive field size is on the order of a fewminutes of arc. At the outer edge of the macula, the receptive field size is aslarge as 3 to 5 degrees. The majority of the ganglion cells are optimally drivenby a center-surround stimulus, or small circular of light at one intensity (orcolor) surrounded by an annulus of light at another intensity (or color). Cellsof the M pathway, representing only 8% of the RGC, are best driven by aluminance difference between the center and surround. Two possible lumi-nance differences are possible: a well-lit circle surrounded by a dark annulus(ON ganglion cell) or a dark circle surrounded by a well-lit annulus (OFFganglion cell). Cells of the P pathway, representing 80% of the RGC, are bestdriven by a color difference in the center and surround. The two types ofoptimal differences are green vs. red and blue vs. yellow. The remaining 12%of the retinal ganglion cells do not fall into the M or P classification and haveuncharacterized function.

The receptive field characteristics of bipolar cells are very similar to thoseof the RGC. In response to the reduction of neurotransmitter release byphotoreceptors with increased illumination, bipolar cells either depolarizeor hyperpolarize; hence, the two classes of bipolar cells are ON center andOFF center. The opposing surround is thought to result from lateral inhibi-tion by horizontal cells. The similarity between bipolar cell and RGC recep-tive field properties is thought to be the result of RGC receiving exclusivelyexcitatory input from a small number of bipolar cells.



11.2.2 Organization of the optic nerve and tract

Together, the optic nerve and tract cross along the ventral surface of thebrain. From the retina to the optic chiasm (the intersection of the nerves fromthe left and right eyes), the nerve is called the optic nerve (ON) and, from thechiasm to the subcortical targets, it is called the optic tract (OT). In humans,both the nerve and tract have a diameter of approximately 3 mm. The lengthof the optic nerve is approximately 50 mm, and the tract is approximately30 mm. Neither the nerve nor the tract has neuronal cell bodies. Instead,each carries the approximately 1,200,000 axons from the retinal ganglion cellsof each eye.3 (In comparison, the auditory nerve has around 30,000 fibers.)The optic nerve contains the axons of the visual fields nasal and temporalto the fovea for a single eye. At the optic chiasm, the fibers are reorganizedso that the optic tract contains almost exclusively the axons representing thecontralateral visual hemifield. In addition to the RGC fibers, the centralartery and vein of the retina pass through the optic nerves, approximatelyin the middle of the nerve.

Indications suggest that the fibers of the optic nerve are visuotopicallyorganized with the upper retina (lower visual field) represented along thedorsal side of the nerve, the central retina along the lateral side, and nasalvisual field along the medial size;4,5 however, this visuotopic organizationappears to vary along the length of the nerve. The distribution of P and Mfibers within the nerve and tract is unclear at this time, but, as the M fibershave twice the conduction velocity, they likely have a larger diameter. Asthe optic nerve and tract are axonal extensions of the retinal ganglion cells,they have the same receptive field characteristics.

11.2.3 Anatomy and physiology of subcortical structures

The axons of the RGC target three subcortical structures, the superior colli-culus, the pretectum, and the lateral geniculate nucleus (LGN), with the major-ity (90%) targeting the LGN. The superior colliculus and the pretectum arelocated on the roof of the midbrain and are associated with saccadic eyemovements and pupillary reflexes, respectively. As blindness results whenthe LGN is damaged but these two structures are preserved, neither structureis well suited for a vision neuroprosthesis.2 The LGN is a small structure(approximately 7

× 7 mm by 2 mm deep) located on the ventral side of thethalamus. The neurons of the LGN are often considered relay neurons,merely receiving input from the RGC and passing the same on to cortex, butthis is an overly simplistic view of the LGN. The LGN has six independentlyacting lamina arranged as six upside down U’s stacked in the ventral–dorsalaxis. Each lamina receives input from only one of the P or M pathways andonly one of the contralateral or ipsilateral eyes. Due to the much higherproportion of P neurons, four laminae are dedicated to this pathway. Eachlamina is visuotopically organized and the visuotopic maps are registeredbetween lamina so that all neurons in a ventral–dorsal path have receptive



fields in the same region of visual space. Almost half of the neurons of theLGN (representing nearly half of the area of the LGN) have receptive fieldsin the foveal or surrounding region. Although a neuroprosthesis could takeadvantage of this unique neural organization, the LGN generally has notbeen considered a potential site due to the surgical difficulties in its exposure.

11.2.4 Visual cortex anatomy and physiology

The final stage of the visual pathway is the visual cortex. Even within thecortex, visual processing continues to be a hierarchical operation with theM and the P pathways segregated. Of the LGN axons projecting to visualcortex, almost all project to the gyri lining the calcarine fissure, a largesulcus running anterior–posterior along the medial surface, just anteriorto the occipital pole. This 2500- to 3200-mm2 region is known as primaryvisual cortex or visual area 1 (V1) to indicate that it is the first region ofvisual processing at the cortical level. The region is also called striate cortex,due to its unique laminar appearance with certain histological stains, andArea 17, from Brodmann’s classification of the regions of cerebral cortex.The regions of cortex subserving visual processing that surround V1,including the lateral surface of cortex anterior to the occipital pole, arecalled visual areas 2, 3, 4, and 5 (V2, V3, V4, and V5), indicating somewhatthe hierarchy of visual processing.

All of these cortical areas have a similar laminar structure, consisting ofsix neural laminae arranged in planes from the pia mater to white matter,with a total thickness of approximately 2 mm in humans. Visual processingis performed both in a columnar manner, with information passing betweenneurons in a column from pia mater to white matter, and in a horizontalmethod, where information is integrated across a number of columns. Gen-erally, the fourth lamina from the pia mater (layer 4) receives input from thepreceding stage in visual processing, layers 2 and 3 project to the next stagein processing, and layer 6 projects back to the preceding stage in processing.In V1, layer 4 is further subdivided into four sublaminae, 4A, 4B, 4C

α, and4C

β. The latter two sublaminae receive LGN input from the M and P path-ways, respectively. Consistent with the general architecture of cortical lam-ina, layer 6 in V1 projects back to the preceding stage in processing, the LGNin this case. Each of the areas of visual cortex is visuotopically organized.Consistent with the columnar processing of information, the visuotopic mapsare registered between lamina so that all neurons in a column from pia materto white matter have receptive fields in the same region of visual space.Almost half of the neurons in any visual area (representing nearly half thesize of the region) have receptive fields in the foveal or surrounding region.The foveal region of visual space is represented at the posterior part of V1.

With the exception of some neurons of layers 4Cα, and 4Cβ, which haveLGN-like receptive fields, the receptive field characteristics of V1 neuronsare much richer than those of the preceding stages of processing. Here, thereceptive fields continue to have subregions preferring illumination (ON



regions) and lack of illumination (OFF regions) but the subregions are elon-gated, more of a hotdog in a bun shape than circular with an annularsurround. The elongation of the subregions leads to a preference for bars ofparticular orientation, resulting in the new receptive field characteristic oforientation preference. The segregation of the retinal input from each eye,so carefully preserved in the laminar structure of the LGN, becomes lessdistinct at the cortical level. Here, a neuron may be best driven by thecontralateral eye or ipsilateral eye, or be equally driven by both eyes. Thisleads to the receptive field characteristic of ocular dominance, the measureof which eye is the dominant driving source. Similar to visuotopy, nearbyneurons tend to prefer the same orientation and receive input from the sameeye, observed both when examining a column from pia to white matter andwhen traversing across cortex. Further, when traversing across cortex, onetends to observe slow changes in either the preferred orientation or preferredeye, or both. A complete sequence of orientations (0 to 180 degrees) and eyepreferences (contralateral to ipsilateral and back to contralateral) occurs overapproximately a 1-mm2 area of cortex. Although it is tempting to map thechanges in orientation and eye preferences as a square 1

× 1-mm grid over-laying the visuotopic map, recent studies have suggested that the prior twomaps have a much more complex organization.6 These studies have shownthe existence of regions where the preferred orientation (or eye dominance)rapidly changes and other regions where the same orientation (or eye dom-inance) is preferred for a few millimeters.

After V1, the organization of the visual pathway becomes more complexand the optimal visual stimuli become less clear. Throughout the visualpathway to V1, the M and P pathways pass through the same neural struc-tures but target different lamina in order to remain distinct. In V2, thepathways begin to diverge with the pathways targeting neighboring hori-zontal subdivisions of V2. Past V2, the pathways completely diverge. The Ppathway heads toward posterior parietal cortex and principally targetsvisual areas on the dorsal aspect of occipital cortex. The M pathway headstoward inferior temporal cortex and principally targets visual areas on theventral aspect of occipital cortex. Further, the simple hierarchical organiza-tion is replaced with a more complex connectivity as can be seen in the blockdiagram of the visual area interconnections shown in Figure 11.4.7,8 Finally,concurrent with higher level feature extraction, the receptive field character-istics V2 neurons and those of subsequent visual areas become much moreintricate. For example, some neurons of inferior temporal cortex are mostresponsive to features that appear like a face.9 Due to complexity of theinterconnectivity and the receptive field characteristics, visual areas beyondV1 have not been proposed as sites for vision prostheses.

11.2.5 Pathologies leading to blindness

The leading causes of untreatable blindness are age-related macular degenera-tion (AMD), retinitis pigmentosa (RP), accidents, and cancers. Additionally,



blindness can be caused by glaucoma and diabetic retinopathy, but thesetwo pathologies currently are treatable and blindness can be prevented iftreated at an early stage.

Age-related macular degeneration is pathology of the retina character-ized by the progressive loss of central vision, generally occurring in theelderly. It is the leading cause of blindness in developed countries, with anestimated 2,000,000 individuals having some form of AMD in the UnitedStates. For unknown reasons, the pigmented epithelium of the retina degen-erates, leading to a subsequent degeneration of the photoreceptors and fluidleakage into the neural retina. As this occurs in the macular region, centralvision is compromised, leading to the loss of the ability to distinguish finedetail. Retinitis pigmentosa is an inherited pathology of the retina charac-terized by the loss of peripheral vision and night vision. An estimated 100,000individuals are affected by RP in the United States. For unknown reasons,the rod photoreceptors degenerate. As these are the principal photoreceptors

Figure 11.4 Schematic representation of the hierarchical processing of informationalong the visual pathway. The two principal pathways (M for magnocelluar pathwayand P for parvocellular pathway) are represented by thick black and thick dark-grayarrows, respectively. These pathways are thought to represent the location (where)and the type (what) of objects in visual space, respectively. Beyond V1, the hierar-chical organization becomes somewhat less clear, suggesting that these regions wouldnot be prime candidates for visual neuroprostheses. (Adapted from Van Essen.7)

Retina

LGN

V1

V2

M P Other

M P SC

4Cα 4Cβ

Other Layers

Thin Stripe Interstripe

V3 VP

PIP V3A

MIP PO MT V4t V4

Other Cortex

MDP

Thick Stripe



outside the macula, peripheral vision is lost. Further, as these photoreceptorsare functional only in dim light, night vision is compromised. Individualswith advanced RP have tunnel vision, only seeing objects with the centrallylocated cone photoreceptors. Despite the degradation of the outer nuclearlayer in both of these retinal forms of blindness, there are reports that innernuclear and retinal ganglion layers are partially preserved, having 80 and30% of the neurons found in a normal eye, respectively.10–12 On the contrary,another group has reported more significant cell loss in transgenetic animalmodels of RP.13 Further, this group reports that the remnant neurons becomeso disorganized and dislocated within the neural retina that the laminarstructure of the retina is not apparent in histological sections.

For other sources of blindness, typically accidents and cancer, either theeye is lost or the visual pathway is disrupted downstream from the retina.In either case, the earliest site in the visual pathway that vision prostheseswould be possible is the LGN.

11.2.6 Concluding remarks

Of the neural structures described previously, only the photoreceptor layer ofthe retina, ganglion cell layer of the retina, the optic nerve, and the cerebralcortical region V1 have been proposed as potential sites for a vision neuro-prosthesis for restoration of function in the blind. These sites principally havebeen proposed due to the relative ease of surgical access, with perhaps theexception of the optic nerve, and the suggestion that the pathology leading toblindness leaves the remnant neurons somewhat functional.13 As will be seenin the subsequent sections, some success has been achieved for all four poten-tial sites. However, due to the need to use “heavy-handed” methods in orderto evoke a useful percept, it is likely that only a crude sense of vision will bepossible in the near term. Nevertheless, using the progress in cochlear neuro-prostheses as an example, once an even partially useful vision neuroprosthesisbecomes available steady improvements will occur and a more useful senseof artificial vision will result. When these second and subsequent generationvision prostheses appear, it is likely they will be specifically designed to takebetter advantage of the characteristics of the underlying neural system.

11.3 Elements of a visual neuroprosthesisIn order for a visual neuroprosthesis to become an accepted therapeuticapproach to sight restoration, it is clear that those who will utilize thistechnology must realize that they will not be receiving sight like those withnormal vision. It is also clear that a clinically acceptable system must bevirtually invisible. This means that the components must be integrated intonormal systems typically worn by individuals such as eyeglasses and anexternal package not much larger than a pocket organizer. While such invis-ibility would be the eventual goal of a commercial system, first-generationexperimental systems are not expected to be so constrained.



A visual neuroprosthesis will be a complex system containing a num-ber of interconnected elements. A block diagram of the elements that avisual prosthesis will likely contain is shown in Figure 11.5. It must containthe following components: a video encoder to capture the visual field infront of the user of the system, signal-processing electronics to transformthe video image into a set of discrete signals that can be used to controlthe injection of current through the neural interface, some mechanism bywhich these processed signals and electrical power can be delivered in awireless fashion to the implanted neural interface, stimulator electroniccircuitry to control the currents injected through each electrode in theneural interface, and the neural interface that evokes the neuronal excita-tion. As this minimum set of components will be present in some form ina retinal, optic nerve, or a cortical visual prosthesis, we will expand uponthese elements in the following sections.

11.3.1 Video encoder

The electrical currents that will be injected at some site into the visualpathways must be related to the visual scene in front of the blind subjectusing the visual prosthesis. Thus, the first key element in the prosthesis isthe video encoder (this element would mimic the lost function of the pho-toreceptors in the retina and transform visible images into electrical images).For cortical or optic-nerve-based visual neuroprostheses, this system couldbe a photodiode array, a dedicated CCD (charge-coupled device) array, aconventional video camera, or a miniaturized video camera, mounted withina pair of eyeglasses worn by the subject (to achieve invisibility of this device).For a retinal-based visual neuroprosthesis, the encoder could be integratedinto the neural interface and, therefore, reside within the plane of the retina.This would have the obvious advantage that the optics of the eye could be

Figure 11.5 Schematic representation of the elements of a visual neuroprosthesis.With the exception of retinal neuroprostheses, all of the devices reviewed in thischapter consist of the items illustrated. Retinal devices, due to their location, mightnot require telemetry and use very simplistic signal-processing and stimulator elec-tronics. Further, these devices derive their power from ambient light. The videoencoder essentially replicates the function of the photoreceptors, that is, transformingthe visual signal into an electrical signal. The signal-processing electronics performnecessary filtering and remapping. As external electronics cannot be as efficient asthe biology, which has been tuned through millions of years of evolution, externalpower may be necessary. The stimulator electronics convert the video signals intoelectrical signals that more effectively stimulate neurons. The interface to the neuronsis provided by an array of electrodes.

VideoEncoder -

'TV camera'

SignalProcessingElectronics

Power andVideoSignal

Telemetry

StimulatorElectronics

NeuralInterface -'Electrode

Array'



used to project an image of the world onto the encoder. Thus, the acquisitionof visual information could function more naturally, without the use of headmovements for image acquisition (normal eye movements could providepart of this function).

The spatial resolution of early generation retinal, optic nerve, or corticalvisual neuroprosthetic systems is expected to be quite low because of thelimited numbers of electrodes in the neural interface. Thus, just about anyconventional video camera could provide adequate spatial resolution for alaboratory-based system. As the temporal resolution of the human visualsystem is relatively low (on the order of 30 Hz), conventional, inexpensivevideo cameras can easily provide adequate spatial and temporal resolutionfor a laboratory experimental system. For a clinical device, miniaturizedvideo cameras are already commercially available. Shown in Figure 11.6 areglasses that contain a miniaturized camera in the bridge of the glasses. Thevery small aperture of the camera (1 mm) makes it virtually invisible andprovides for a very large depth of field without the use of additional lenses.The camera provides a color NTSC signal and allows for encoding imagesat distance, and for encoding of text when the text is brought close to theglasses. Also shown in the figure are examples of images of text (font size12) and an office encoded with this camera and sent to a black and whitemonitor (the photographs were taken off the monitor screen).

11.3.2 Signal Processing

As discussed earlier in this chapter, the foundation upon which a visualprosthesis will be built is the notion that the visual system is organized intoa hierarchical sequence of maps. One of the most relevant maps is calledvisuotopy: points in space that are close to each other excite neurons at variouslevels in the visual pathways that are also close to each other. If this visuo-topic mapping were perfectly conformal and linear, then a high-qualityneural “image” of the visual space would be found at each level in the visualpathways. This is not the case, however. The visuotopic map is only confor-

Figure 11.6 Example of video encoder. The first panel illustrates glasses that containa miniaturized camera in the bridge of the glasses and the associated electronics thateasily could fit into a pocket. The electronics provide a standard color NTSC signal.The second panel illustrates printed text as viewed through the camera, and the thirdpanel shows one of the authors of this chapter as viewed through the camera. Bothpanels have been converted to gray scale.



mal when considered from a low-resolution perspective. If one studies thevisuotopic organization with high spatial resolution techniques, it becomesclear that the mapping is somewhat locally random. Recent work by Warrenet al.14 have shown that one can use the relative location of a neuron in thecortex to predict where spots of light projected in visual space will excitethe neuron. However, this can only be done with an accuracy of about 1/2degree of visual angle. This uncertainty reflects the nonconformal nature ofthis visuotopic organization.

Signal-processing electronics must accomplish a number of tasks. First,the electronics must transform the signal coming out of the video encoderinto a set of discrete signals, one for each electrode in the neural interface.Next, the encoder must be able to adapt the incoming light signals into arange of stimulus levels that is appropriate for the neurons that are beingstimulated; that is, the range of stimulus levels must be the same regardlessof the level of ambient illumination (a bright, sunny afternoon or a dimly litrestaurant). This will require an automatic gain control circuit that willduplicate the adaptation properties of the human photoreceptors. The signal-processing electronics could also perform image compression, somewhat likethat achieved with the MPEG techniques used to compress audio and visualimages in digital cameras and DVD video disks. Finally, the signal-process-ing electronics may have to accomplish remapping of the visual image suchthat a vertical line in visual space evokes the percept of a vertical line. Thedegree to which this remapping may be required will depend upon thedegree of randomness in the visuotopic organization at the implant site andthe degree of plasticity that remains in the subject’s visual system. Thisproblem of remapping of visual space into appropriate signals is beingpursued by the Eckmiller laboratory in Germany.15 While this level of signalprocessing appears highly complex, recent developments in integrated cir-cuit technology will make this a complex but achievable task.

11.3.3 Telemetry and power interface

In order to achieve the challenge of developing an invisible clinical visualneuroprosthesis, it will be necessary to use a wireless link by which powerand video signals are delivered to the implanted neural interface. The designconstraints of such a wireless link are such that they are only recently becom-ing possible due to recent advances in integrated circuit technology. Thewireless link could be mediated by radiofrequency or by light, and recentwork on integrated optical systems and cellphone technologies are makingthese approaches tenable. The telemetry link should be bidirectional so theimplanted electronics can inform the external electronics of the need for moreor less power. In order to minimize excessive power requirements of thetransmitting electronics, it is expected that the distance between the trans-mitter and receiver will be on the order of a centimeter. This suggests thatthe receiving coil will likely be implanted under the scalp (for a corticalprosthesis) or within the eye for a retinal based system. The coupling of



power between transmitters and receivers can be achieved with very highefficiencies over such short distances. One possible design would be the useof transmitting coils built into the image-encoding eyeglasses. Such a designwill have the advantage that registering the transmitting coil over theimplanted receiving coil can be readily accomplished and will not varysignificantly during the course of normal daily activities of the wearer. Fur-ther, the external electronics would consist only of the eyeglasses, signal-processing electronics, and the telemetry circuitry.

In order to ensure that the telemetry system does not deliver damaginglevels of heat and radiation to the tissues near the implanted electronics, thefrequency of the radiated power must be limited and the efficiencies of theimplanted electronics must be very high. Further, the implanted receivermust be very small and have ultra-high reliability, considering that theimplanted electronics will be immersed in the corrosive environment of thehuman body, possibly for decades. Finally, the bandwidth of the telemeteredvideo signals will depend upon a number of factors. As the number ofelectrodes in the implanted neural interface increases, the bandwidth of thetransmitted signals increases. If the signal-processing electronics use cleverencoding strategies, the bandwidth of the video signals can be significantlyreduced. It is clear that all these design considerations present the designengineer with very significant challenges.

An experimental wireless power and video signal telemetry link hasbeen designed and fabricated by Dr. Phillip Troyk at the Indiana Institute ofTechnology. This system is intended to be used in a cortical visual prosthesis.The system has a modular design so that failure of any individual compo-nents will not cause failure of other modules. The system has also beendesigned to consume little power and to facilitate the interconnection ofneural interfaces.

11.3.4 Neural stimulator

The video signals will have to be processed to make them compatible with,and appropriate for, stimulation of neurons at some level in the visual path-ways. Thus, the next major element in the visual prosthesis is the neuralstimulator. As we have discussed earlier, an experimental visual prosthesissystem may use external stimulator electronics, which would offer theadvantage that commercially available electronic stimulators could be used.However, clinical systems will likely require that the stimulator electronicsbe implanted near to the neural interface. The stimulator will most likely bea digital device, which will facilitate its development and minimize powerdissipation by the device. Thus, the stimulator will likely contain on-chipmemory locations, with one location dedicated to each electrode in the neuralinterface. These memory locations will be updated via the telemetry link asthe image being encoded changes. The considerations associated with thedesign of a neural stimulator have been discussed in a paper by Jones et al.,16

and this design has been implemented in a VLSI (very-large-scale integra-



tion) chip. The problems of hermetically sealing the electronics, however,represent one of the more formidable challenges for the design engineer.

11.3.5 Neural interface

Any visual prostheses will require an interface between the nervous systemand internal and external electronics, a so-called brain–computer interface.This interface should be thought of as an active element that transduceselectronic currents that flow in the human-engineered devices that areconnected to the device into ionic currents that flow in the human body.While most materials accomplish this transduction to a limited degree,certain materials do this relatively well. However, some of these materials,such as silver, are toxic to the neurons in the immediate vicinity. Platinumis a highly biocompatible material but it is not particularly effective intransducing electrons into ions. Oxidized iridium, on the other hand, isnot only biocompatible but is also an excellent electronic-to-ionic trans-ducer. It is the material of choice today and is generally being used in mostimplantable neural interfaces.17

In other human-engineered implant systems, the interface between theliving and non-living has generally been difficult to develop. The humanbody has developed numerous defense systems that have specificallyevolved to ward off intrusions into the body. These biological defense sys-tems are so effective that most devices that are implanted in the body even-tually either become degraded into harmless molecular species or becomesealed off by a fibrotic tissue capsule. The challenge of the bioengineer is tothwart this biological process with the use of materials that the body recog-nizes as benign. While this seems like a straightforward task, the more onedelves into the problem, the more complex it becomes.

11.3.5.1 Physical biocompatibility

11.3.5.1.1 Density of the implant system. Ideally, the implant systemmust appear invisible to the host tissue. The invisibility applies not only tothe chemical nature of the surfaces that come in contact with the host tissuebut also to the physical aspects of the implant. Specifically, if the density ofthe implant is not identical with that of the host tissues, gravity or kinematicaccelerations associated with normal body movements can produce shearforces between the implant and the host tissue. If these shear forces aresufficiently large, the implant can migrate through the tissue, or at leastexperience micromovements with respect to the host tissue. These micro-movements can produce a chronic inflammatory response to the host tissuedue to chronic mechanical irritation of the neurons near the implant. Thisproblem is exacerbated in the visual pathway, where saccadic eye move-ments can produce large accelerations of the eye. Any neural interface thatis implanted within the eye could be subjected to particularly large shearforces if the density of the implant differs significantly from the host tissues.



11.3.5.1.2 Physical barriers. The cells of the nervous system are in ahomeostatic equilibrium with their local environment. Each cell exchangesnutrients and waste products with the extracellular fluid surrounding the cell,and these molecules eventually become transported into or out of the capil-laries that support the tissues. Implanting large structures into the nervoussystem, even if their surfaces are biochemically invisible, could disrupt thisexchange of nutrients and wastes. This aspect of physical biocompatibility isparticularly relevant in various tissues of the visual system such as the retina.The oxygen that nourishes the retina comes from two sources: the choroidaland retinal circulations. Similarly, the carbon dioxide waste is also removedfrom the retina via these two circulations. The choroidal circulation is a highlyvascularized bed that ramifies throughout the choroid (located between thepigment epithelium and the sclera) and, because of its proximity to the pho-toreceptors, provides the major nourishment of the outer retina (photorecep-tors, horizontal cells and bipolar cells). It also nourishes the pigment epithe-lium. Any implant system placed at any point between the choroid and theretina must not act as a significant barrier to the exchange of oxygen andcarbon dioxide between these tissues, or retinal degeneration could be exac-erbated by the implant system. The retinal circulation nourishes the cells inthe inner retina (the ganglion cells, amacrine cells, and, to some degree, thebipolar cells). As the retinal circulation is integrated with the retina itself, itis difficult to imagine how an implant system could be interposed betweenthe retinal circulation and the cells it supports. However, the existence of anymolecular exchange between the vitreous humor and the inner retina couldbe compromised by implant systems placed on the retinal surface. The degreeto which molecular exchange between the pigment epithelium and the retinatakes place and its importance in retinal support are not clear.

11.3.5.1.3 Mechanical compliance of the implant system. The neural tis-sues into which a visual prosthesis will be implanted are relatively compliant,compared with the materials with which the implant system is typicallyconstructed. Further, the tissues into which (or onto which) the implant isinserted typically experience relative movements due to gross movements ofthe body or more subtle displacements due to blood pulsations or to respi-ration. Because of the compliant nature of the host tissues, the tissues can beslightly compressed or distended as a result of these mechanical displace-ments. Clearly, if the implant materials are virtually noncompliant, a mechan-ical compliance mismatch can develop at the interface between the hosttissues and the implant. Polymeric materials are generally more compliantthan the metals and silicon used in many implant systems. It is likely thatfuture generation implant systems may profitably utilize new generations ofpolymeric materials to mitigate this potential compliance mismatch problem.

11.3.5.1.4 Tethering produced by lead wires. While clinical visual pros-theses will eventually be wireless devices, where power and signals will bedelivered across living tissues via some sort of telemetry system, this might



not be the case for first-generation systems. The goal of many first-generationvisual prostheses will be to demonstrate proof of concept, that the systemcan produce functional (but probably very limited) vision. These first-gen-eration systems are likely to have lead wires that run from the neural inter-face that is implanted in the host tissues to some sort of a connector, mountedat some stable site on the body. Even if signals and power are deliveredacross the skin via some form of telemetry, the size of today’s VLSI circuitryis such that it is still likely that some form of lead wire system will still beused between the implanted neural interface and the implanted telemetryand stimulating VLSI circuitry. Thus, it is quite likely that any first-generationvisual prosthesis will have to deal with the problem produced by tetheringfrom these lead wires. The problem of tethering can be best appreciated byconsidering a cortically based visual prosthesis, where an array of electrodesis implanted on or in the visual cortex, and a connector is attached behindthe ear. Even if the mechanical problems outlined above are dealt with, anymotion of the brain with respect to the skull will result in the generation ofrelative motion between the brain and neural interface, which could producepotential chronic inflammation. These problems can be partially mitigatedby various strain-release lead wire architectures and by anchoring of the leadwires to the cortex near the implant site. However, these measures will onlywill reduce, not eliminate, the tethering problems.

11.3.5.2 Biological biocompatibilityWhile the goal of achieving a biological invisibility of the neural interface isdaunting, the problem is mitigated to some degree by two factors: a judiciouschoice of the materials from which the interface is constructed and the well-documented immunologically privileged nature of the nervous system toforeign materials. The implantation of any neural interface, at any site in thenervous system, can be thought of as creating a wound at that site, no matterhow benign. There is an extensive literature on the mechanisms and kineticsof wound healing that go well beyond the scope of this chapter, and thereader who is developing neural interfaces is encouraged to become familiarwith this literature. To summarize for the purposes of this chapter, the natureof the wound healing response will depend upon the nature and extent ofthe wound and the materials from which the wounding device are manu-factured. We will consider below the issue of materials.

Again, the quest for safe implant materials has been pursued for manydecades and has been inspired by the development of implantable joints,pacemakers, catheters, cosmetic applications, and orthopedic devices, toname but a few more prominent devices. There is also an extensive literatureon the biocompatibility of a wide spectrum of materials, but implant systemscurrently under development use a subset of these materials: silicone, silicon,platinum, iridium, gold, polyamide, parylene, Teflon, and PMMA(poly[methyl methacrylate]). These materials are used for various purposesin the construction of a neural interface: structural, charge carrying, elec-tronic-to-ionic transduction, and insulation. These materials have been used



because they provide the architectural and functional properties from whicha neural interface can be constructed and because they are generally regardedas being quite biocompatible.

11.4 Subretinal neuroprosthetic approachAs some of the neural retina, with the exception of the photoreceptor layer,is preserved for both age-related macular degeneration12 and retinitis pig-mentosa,10,11 a reasonable approach to the restoration of vision under thesepathologies is the replication of the function of the photoreceptors. (Anothergroup, however, has reported more severe retinal degeneration in animalmodels.13) From a simplistic engineering viewpoint, the photoreceptors ofthe outer nuclear layer appear as a tightly packed array of light-to-currenttransducers, a function readily replicated by arrays of semiconductor pho-totransducers. Two research groups have focused on replacing the functionof photoreceptors with an engineered system: (1) the Optobionics group, ledby Alan Y. Chow and Vincent Y. Chow; and (2) a consortium centered inTubingen, Germany, led by E. Zrenner.18 In the general sense, both groupshave proposed a similar approach. An array of phototransducing elementsis placed in the subretinal space, between the pigmented epithelium and thedegenerating outer nuclear layer. Each element (or subunit) consists of asemiconductor-based photodiode and an electrode. Light striking the pho-todiode causes a photocurrent that flows between the electrode and a refer-ence electrode, the entire back surface of the array. The result is a voltagegradient that is intended to stimulate the dendrites of the bipolar cells in theouter plexiform layer. Because the stimulating electrodes are placed in thesubretinal space, this tactic has been termed the subretinal approach.

The scheme is tantalizingly simple, requiring no external power or con-trol signals. Further, the fabrication of the photodiodes can be tailored toprovide either positive or negative current in response to illumination, withthe intention of mimicking the operation of ON and OFF bipolar cells. Bothgroups have suggested that, in theory, no signal processing would be nec-essary, as they claim that the entire neural network of the visual pathwaywould be stimulated in a physiologically normal manner. The success ofsuch an approach depends upon three assumptions. First, the bipolar cellsof the dysfunctional retina persist and function in a somewhat physiologi-cally normal manner. Second, the photodiodes can produce sufficient currentunder normal illumination levels. Third, the electrodes can be placed in closeenough proximity to the bipolar cells such that the induced photocurrentexcites the bipolar cells. As will be discussed in this section, the validity ofthese assumptions, particularly the second, remains very controversial.

11.4.1 Optobionics group

The Optobionics Corporation, a privately held company located in Wheaton,IL, has developed the Artificial Silicon Retina (ASR), which entered into



clinical trials of its safety in the summer of 2000. (Optobionics, ArtificialSilicon Retina, and ASR are trademarks of the Optobionics Corporation.)This device represents the commercialization of the scientific research ofChow and Chow, two brothers who continue to provide the scientific lead-ership of the company. Much of their research has centered on the safetyand efficacy of a subretinal neuroprosthetic device.

In their early research, Chow and Chow showed that a simple combi-nation of an electrode placed in the subretinal space and a photodiode couldbecome the basis of a visual prosthesis.19 Using a single, very large goldelectrode (36 mm2 surface area or about twice the size of the macula inhumans) acutely implanted into the subretinal space of an anesthetizedrabbit, they observed that illumination of a photodiode connected to theelectrode induced cortical activity. The cortical activity, measured by elec-trodes placed on the scalp of the rabbit, had amplitude and temporal char-acteristics similar to the visually evoked potential measured in the unoper-ated eye. Further, the magnitude of the cortical-evoked potential could bemodulated by adjusting the intensity of the stimulus or the size of thephotodiode. With a charge density of 100-nC/cm2, resulting from 12-kluxillumination (similar to a bright, sunny day) on a 29-mm2 photodiode, themagnitude of evoked potential was similar to the visually evoked potentialmeasured in the unoperated eye.

Having shown a proof of concept, their research moved into the designand test of an array of photodiodes and electrodes manufactured by standardsemiconductor methods. In a series of articles, they have reported on thebiological response to an implanted array and the ability to evoke neuralactivity with the array.20–25 The test ASR has remained relatively unchangedexcept in its thickness throughout these studies. At the macroscopic scale,the device is a 1- to 3-mm-diameter disk (0.8- to 7-mm2 area), initially being250

µm thick and reduced to 50 µm thickness in the most recent report. Atthe microscopic scale, the test ASR is an array of photodiode and electrodeelements, each element being approximately 20 × 20 µm by the thickness ofthe disk. Insulating moats (10-µm width) separate neighboring elements,resulting in a density of approximately 1100 elements/mm2. Each elementconsists of a photodiode created by variously doped layers of silicon, sand-wiched between two layers of gold. On the front side of the array (sideclosest to the light), the gold electrode is etched to the same dimension ofthe photodiode. Light reaching the photodiode passes through the gold layer,which is relatively transparent due to its thinness. The gold on the backsideof the array forms a single sheet and acts as a return electrode. As gold doesnot readily adhere to silicon, a chromium adhesion layer is place betweenthe gold and the silicon. The photodiode is responsive to light in the 500- to1100-nm wavelength range, which covers most of the typical range inhumans (360 to 830 nm) plus adding an infrared (IR) response.

In these reports, histological examinations and the electroretinogram(ERG) were used to assay the tissue response to the implant. In chronicallyimplanted cats, they observed a significant loss of cells in the outer nuclear



layer but insignificant changes in the cell density for both the inner nuclearlayer and retinal ganglion layers.24,25 Additionally, the appearance of giantcells indicated an inflammatory response to the implanted device. As AMDand RP are characterized by the loss of the outer nuclear layer, its loss inthese animal experiments is considered inconsequential. Outside the site ofimplant, the retina had a normal laminar arrangement. The lack of a signif-icant immune response to the implant is not surprising as the device isconstructed from well-tolerated materials. Photic stimulation of theimplanted eye resulted in an ERG similar to the unimplanted eye. This alsois not an unexpected result, as the implant is relatively small and implantmaterials are well tolerated. If the device led to an out-of-control immuneresponse, it likely would have been apparent in the histology long beforeaffecting the ERG.

The ERG response for IR stimulation was used to verify the implantcontinued to be functional.24 The ASR is much more sensitive to IR stimula-tion than the native photoreceptors. These data indicated that the test ASRcontinued to function as a photodiode array for up to 11 months, the longesttime duration reported. However, the magnitude of the ERG took 1 to 2months to reach its maximum and began to degrade after 4 months. Fromthe data provided, the ERG response had a magnitude of up to 75 µV, andthe sign of the response depended on the type of photodiode utilized (neg-ative or positive current generating). The latter finding is a clear indicationthat this ERG response is simply a stimulus artifact. If the ERG representedneural activity and the two polarities were equally effective at elicitingactivity, the two ERGs would be equal and similarly signed. If one polaritywere more effective, the more likely case, the ERG would appear when theillumination is initiated for one polarity and when the illumination is termi-nated for the other polarity. At the end of the study, the arrays were explantedand examined. It was found that the gold electrode material was degradedor absent for active devices, suggesting that the gold electrodes were beingdissolved into the tissue by electrical stimulation. Alternative electrode mate-rials, platinum and iridium oxide, have been investigated but no specificrecommendations are available at this time.23

In addition to animal experiments, Optobionics has initiated FDA-approved clinic trials to investigate the safety of the ASR in humans. Thedevice currently in clinical trials is a 2-mm-diameter disk of 25 µm thickness.Each photodiode and electrode element has the same 20 × 20-µm dimensionswith 10-µm moats separating adjacent elements. Given the approximately3500 elements on the device, the total photodiode (and electrode) surfacearea is around 1.4 mm2. In comparison, the devices used in their animalexperiments, which had a complete lack of neural response for what shouldbe effective stimulation, had a photodiode area of up to 7 mm2. The onlydevice where a neural response was assuredly observed had a surface areaof around 30 mm2 and then under aggressive stimulation.19 The results ofthese clinical trials are only available in abstract form at this time.26 Theimplants are reported to be well tolerated and have not resulted in infection,



inflammation, or retinal detachment. Additionally, the implants are reportedto have improved the recipient’s visual function; however, these results areonly available as recipient testimonials, not controlled scientific studies. Caremust be taken to distinguish the improvement due to the activity of thedevice and the improvement that occurs simply by the introduction of thedevice. It has been established that a transient, functional rescue of a dys-functional retina occurs in animal models as the result of surgical interven-tion into the subretinal space, even if the intervention does nothing morethan expose the subretinal space. Nevertheless, the test of the biocompati-bility of the materials and the development of the implant techniques makethe ASR a groundbreaking device.

11.4.2 Southern German group

The southern German consortium, centered in Tubingen, has primarilyinvestigated various engineering design issues for a subretinal implant. Theirtest device, called a multiphotodiode array (MPDA), is very similar to the onedescribed in the previous section.27 The individual photodiode and electrodeelements are the same size, 20 × 20 µm, but the moats between elements areonly 5 µm wide, leading to a density of 1600 elements/mm2. Instead ofcovering the entire inner surface of the photodiode, the electrode of eachelement is an 8 × 8-µm square located at the center of the photodiode. Thesmaller electrode size allows more incident light to reach the photodiodeand the electrode layer to be thicker. Gold, iridium, and titanium nitridehave been investigated as electrode materials. The remaining inner surfaceof the element is insulated with silicon oxide, which is transparent to light.Details of the silicon doping used to create the photodiodes are not available.

This group took a very different approach, and arguably more usefulapproach, to the proof of concept. Using retinas isolated either from newlyhatched chickens or Royal College of Surgeons (RCS) rats, they have shownthat photic stimulation of the MPDA modulates the activity of retinalganglion cells.27,28 This method allows a much finer assay of the impact ofelectrical stimulation than the ERG and cortical potentials, which measurethe mass action of large groups of neurons. In these experiments, a retinais removed from an animal and is sandwiched between the MPDA and amultisite recording electrode, with the ganglion cell layer closest to therecording electrode. To activate the MPDA, a spot of light of known sizeand illuminance is shined through the recording electrode and retina. Theirrecording electrode, consisting of an array of small metal electrodessparsely distributed across a glass slide, is transparent to light. With thisdevice, one can pick up the spiking activity of individual retinal ganglioncells which is easily distinguishable from the graded potentials of thephotoreceptors and bipolar cells. To assure the ganglion cell activity is theresult of the MPDA stimulation, the photoreceptor layer was damaged forthese experiments. In the case of the chick retina, this was done by anunspecified method. In the case of the RCS rat, it is not necessary to damage



the photoreceptor layer; this strain of rats is a well-known model forretinitis pigmentosa and, in mature individuals, has almost no residualphotoreceptors. Tests with sham devices were used to verify the resultswere not due to residual photoreceptors.

In both of these species, the activity of individual ganglion cells couldbe modulated by light, presumably by the activity of the MPDA. However,very strong illuminance was necessary, between 10 and 100 klux (outside ona bright day). Further, as the diameter of the light was reduced, it becamemore difficult to evoke any ganglion cell activity. For example, with a 70-klux, 0.250-mm-diameter spot of light illuminating approximately 80 pho-todiodes, a barely perceptible change in the activity of one ganglion cellcould be detected but only through repeated trials. When only a few pho-todiodes were illuminated, ganglion cell activity could not be inducedregardless of the illuminance level. From these results, this group has con-cluded that a passive photodiode array cannot provide a useful sense ofvision under normal lighting conditions.29

This group has extensively studied the biological tolerance to the materialsbeing implanted and how coatings can enhance the tolerance.30 Using disso-ciated retinal cells from a neonatal rat, the cell adhesion to various coated anduncoated implant materials were tested. To enhance adhesion, three differentcoating materials were examined, poly-L-lysine, poly-D-lysine, and laminin.When applied to an iridium surface, all three coating materials significantlyimproved the cell adhesion over an uncoated surface. Of these, the best adhe-sion was seen with poly-L-lysine, with approximately 75% of cells adhering.Therefore, all implant materials were coated with poly-L-lysine prior to testing.Of the implant materials tested (silicon, silicon oxide, silicon nitride, iridium,and titanium nitride), only titanium nitride exhibited a significantly poor celladhesion. In comparison to the control (poly-L-lysine on glass), only 30% ofthe cells adhered after 28 days of culturing. This poor adhesion was notconsidered to be due to release of toxic materials and subsequent cell death,as cells readily adhered to a poly-L-lysine-coated glass coverslip introducedinto the same Petri dish. As this group had proposed using a titanium nitrideelectrode, which they report has the greatest safe charge injection capacity, thiswas a discouraging finding. One of their electrode materials, gold, was nottested. The geometry of the MPDA does not appear to affect cell adhesion.With an iridium electrode, the adhesion was the same as the control. Whenusing a titanium nitride electrode, the cell adhesion was improved but stillwas poor in comparison to the control. It is unclear whether the MPDA wasactive or inactive during these tests. An interesting but apparently untestedcomparison could be made between an active and inactive MPDA, therebyinvestigating whether the electrodes release toxic substances when active.

In addition to these in vitro tests, in vivo tests of the long-term functionand biological tolerance of the MPDA have been performed in both rabbit andpigmented rat (Long Evans).27,28 The ERG response to infrared stimulation (300mW/cm2 at 940 nm) was used to validate the survival of the implant. Thedetails of the time course of the ERG magnitude are not available but are



reported to be stable for up to 20 months, the longest time duration tested.Concurring with their excised retina experiments, there was no sign that IRstimulation resulted in activation of the inner retina. Without activation of theinner retina, the MPDA cannot provide functional restoration of sight. Thewhite-light-induced ERG and postmortem histological examination assayedthe biocompatibility of the implant but only in the rat model. A reduction ofthe ERG response was reported in three of five rats but the magnitude andtime course were not detailed. Four months after implantation, the retinaabove the implant site exhibited a nearly complete loss of the outer nuclearlayer as well as atrophy in the outer plexiform layer. As mentioned earlier, theloss of the photoreceptor layer is considered inconsequential. The thickness ofthe inner nuclear layer and inner plexiform layer were reported to beunchanged but it is unclear what it was compared against. The cell densitiesin the inner nuclear layer and the ganglion cell layer were similar to thoseobserved away from the implant site. However, the stain utilized was insen-sitive to cell type and the authors clearly concede that its is possible, butunlikely, that all neurons have been replaced by other cell types. Additionally,the excised tissue was tested for glia fibrillary acidic protein (GFAP), a generalmarker expressed by glia in degenerating retina and Muller cells and astro-cytes in the retina. In comparison to controls, GFAP expression was moreprevalent in the implanted retinas. The reason for the increased expressionand its impact are not clear at this time. From these results, this group hasconcluded that a subretinal neuroprosthesis will survive in the long term andis well tolerated. To validate the functional restoration of vision, they haveproposed implanting the MPDA in RCS rats, which have a degenerated pho-toreceptor layer. However, they recognize the lack of a generally accepted andunambiguous method of verifying the restoration of function.

Due to the relatively strong illuminance necessary to induce neuralactivity by the passive MPDA, this group is also activity investigating anactive subretinal device.28,31 Such a device would be very similar to theepiretinal approach but would place the array of stimulating electrodes inthe subretinal space. This approach is still in the proof of concept phase atthis time. In their proof of concept tests, an isolated retina is placed on amultisite stimulating electrode, photoreceptor side closest to the electrodes,and the activity of a retinal ganglion cells is monitored with single micro-electrodes. They have found that ganglion cell activity can be induced witha relatively low median charge of 0.4 nC. However, due to the small size oftheir electrodes (80 µm2), the charge density is 500 µC/cm2, well above thesafe limit for most materials. At this point, an active subretinal device mustbe considered in the investigational phase.

11.4.3 Summary

Having a test device in clinical trials notwithstanding, the subretinalapproach has many unresolved problems. Principal among these is whethera passive device can generate sufficient current to evoke ganglion cell activity



under normal lighting conditions. In their most recent report, the southernGermany group concluded that it was not possible and has moved to anactive device.29 The Optobionics group believes it is possible and has movedto clinical trials of the device’s safety.

In the reports where adequate details are given regarding the nature ofthe illumination, an effect on the subsequent neural pathway occurred onlyfor relatively bright lights over relatively large regions of the test prosthesis.The Optobionics group reported that large, electrically evoked cortical poten-tials could be had with 12-klux illuminance applied to a photodiode havinga 30-mm2 surface area.19 Using a finer assay of subsequent neural activity,retinal ganglion cell activity, the southern Germany group reported that asmall but noticeable modulation could be had with a 70 klux, 250-µm-diameter spot.28 Although the 10- to 100-klux range of illuminance is phys-iologically relevant, this level of illuminance is typically only seen outsidein bright sunlight. More typical ranges of illuminance encountered are 1 to1000 lux. The Illuminating Engineering Society of North America (IESNA)recommends illuminance levels of 260, 520, 700, and 2200 lux for a hotelroom, library reading room, laboratory, and hospital operating room, respec-tively. Further, the lack of a noticeable neural response to IR stimulation castsfurther doubt on the applicability of a passive device. Specifically, becausethe published IR-induced ERG data do have a b-wave component, whichindicates activity in the inner retina, neural activation is unlikely.24,28

Assuming one could develop sufficient current at the commonly encoun-tered illuminance or if one assumes that the device is operational only inbright light, a passive subretinal device has a significant design flaw in thatit cannot adapt to varying light levels. Assuming a linear response, such adevice would be functional over 1 to 2 orders of magnitude of illumination.In contrast, our natural vision functions over 7 orders of magnitude. Nev-ertheless, it can be argued that a functional device that operates with somelimitations would be an improvement over blindness.

Another area of concern is the destruction of the remnant photoreceptorlayer and well as a significant impact on the outer plexiform layer resulting fromthe implant.25,28 These cytological changes are presumed to be the result of thedevice blocking the free flow of nutrients and waste between the neural tissueand its blood supply. Both groups have proposed mesh-like subretinal implants,which would allow free flow, but the fabrication method for such devices is notclear. Unlike the outer retina, the inner retina appears unaffected by the implant.However, the studies to date simply examine the density of cells in each of theinner nuclear layer and retinal ganglion cell layer without examining the pos-sibility of a functional consequence of the implant. Additionally, expression ofGFAP by the glial cells is increased, but the consequence is unclear.

11.5 Epiretinal neuroprosthetic approachAs an alternative to stimulating the remnant bipolar cells in the dysfunctionalretina, other research groups have proposed placing stimulating electrodes



on the inner surface of the retina and stimulating the remnant retinal gan-glion cells. Reports indicate that significant portions of the retinal ganglioncells survive even in the end stage of pathologies such as age-related maculardegeneration12 and retinitis pigmentosa,10,11 giving credence to this approach(however, a significant disorganization of the retina has been reported inanimal models13). The principal research groups investigating this approachare (1) a group led by Humayun and de Juan, which was located at JohnsHopkins University but recently has moved to the University of SouthernCalifornia; (2) a collaboration between MIT and Harvard, led by Rizzo andWyatt; and (3) a consortium centered in Bonn, Germany, initially led byEckmiller and now led by Eysel.32 In the general sense, all three groups haveproposed a similar approach. An array of surface electrodes, attached to theinner surface of the retina between the vitreous humor and the inner limitingmembrane, provides the interface to the neural retina. Driven by signal-processing electronics, patterns of these electrodes are electrically stimulatedto produce a similar pattern of phosphenes, a consequence of the visuotopicorganization of the ganglion cells. In a clinical device, the signal-processingelectronics will be placed within the eye with both the power and the controlof the signal-processing electronics provided by an optical or radio frequencylink. As the stimulating electrodes are placed on the surface of the retina,this tactic has been termed the epiretinal approach.

This approach tries to make use of the advantages of a retinal prosthesis,chief of which is a simple, linear visuotopic organization, while recognizingwhat the other neuroprosthesis groups have long ago recognized, the needfor more power and signal-processing capability than is available from sim-ple phototransducers. The success of this approach depends upon fourassumptions. First, the retinal ganglion cells of the dysfunctional retina per-sist and function in a somewhat physiologically normal manner. Second, theneural interface can be permanently attached to retinal without significantdamage. Third, useful percepts can be had at safe stimulation levels. Fourth,patterned percepts can reliably be created.

11.5.1 Humayun and de Juan group

The Humayun and de Juan group initially began their work as part ofcollaboration between Johns Hopkins University and North Carolina StateUniversity. More recently, they have moved to the University of SouthernCalifornia. Additionally, this group has a commercial relationship with Sec-ond Sight, a privately held company located in Valencia, CA, which is testinga retinal neuroprosthesis in clinical trials. This group has primarily investi-gated the biological aspects of an epiretinal implant, including a large num-ber of experiments in human volunteers.

In their early research, they showed that electrical stimulation of theinner retina results in neural activity of the retina, or a general proof ofconcept.33 These experiments were performed both with isolated bullfrogeyecup preparations and with in vivo rabbit retinas. Using platinum, bipolar



stimulating electrodes, manually placed on the inner retinal surface, theyexcited the tissue with biphasic current pulses of between 50 and 300 µAamplitude, lasting 75 µs per half phase. Each electrode had around 0.13 mm2

surface area and the tips of the bipolar pair were 200 µm apart. A secondbipolar electrode, manually placed between the stimulating electrodes andthe optic disk, sensed the neural response. In response to electrical stimula-tion, they observed a 2-msec long waveshape that had the appearance of anextracellular measured action potential. No response was observed if thestimulating electrodes were placed 1 mm off the retinal surface or if therecording electrodes and stimulating electrodes were placed on oppositesides of the optic disk. They reported that, for current levels as low as 50 µAin the bullfrog and 150 µA in the rabbit, a neural response could be observed.This is equivalent to charge densities of 3 and 9 µC/cm2, respectively. Thisis well within the safe current density levels (<100 µC/cm2) for platinum.17

Their data indicate that the average magnitude of the neural response grewfrom 10 µV (peak to peak) to 20 µV as the current was increased from 150µA to 300 µA in the rabbit preparation. This growth likely indicates thatmore ganglion cells were being excited at larger current levels but their reportclaimed the growth was not significant. In addition to using normal rabbits,they also tested a rabbit model of a degenerate photoreceptor layer, causedby intravenous injection of sodium iodate well before testing. With thispreparation, they were able to observe a neural response but the requisitecurrent level increased to 200 µA.

Having shown that electrical stimulation of the retina can producedlocalized neural activity, this group began investigating two key issues fora clinical device: (1) the possibility of selective stimulation of particularstructures through changes to the stimulus paradigm, and (2) the biocom-patibility of the implant materials, particularly the method of adhering theimplant to the retina. The first area was explored by computation modelingstudies of electrical stimulation of retina.34 In particular, these studies weredirected at distinguishing which neural structure is activated by electricalstimulation, the soma of retinal ganglion cells or passing axons from otherretinal ganglion cells, the latter being between the stimulation source andthe somas. To provide visuotopically organized stimulation, it is preferableto stimulate the somas and not the axons. The results of the study suggestthat there is a penchant for stimulating the soma but the preference wasslight. No specific recommendations for changes in the stimulus paradigmcame from this work.

The biocompatibility issue was investigated with electrode arrays, bothpassive electrodes and actively stimulated electrodes, chronically implantedin animals. In both studies, the array consisted of large platinum disks placedin a silicone matrix. For the passive studies, the entire array was 3 × 5 mmby 1 mm thick. Specific details of the active electrodes are not available. Theresults from the passive implants suggest that the implant materials are welltolerated, indicated by both histological and electrophysiological examina-tion.35 However, the metrics employed solely indicate that the implanted



retina continues to function and does not ensure that the retina under thearray functions normally. No histological comparisons were made betweenthe retina under the implant, approximately 15 mm2 in area, and the remain-der of the retina, over 1000 mm2. In the second set of studies, the electrodeswere stimulated with 0.05 or 0.1 µC/cm2 charge densities, 10 to 12 hours aday for up to 60 days. The stimulation did not cause any gross changes inthe function of the retina but the finer assay of tissue histology has yet to bedone.36 Interestingly, these researchers chose not to utilize a charge densitywhere a pathological tissue response is expected, thereby providing a posi-tive control. In both studies, one or two retinal tacks were used to hold thearrays in place. This group has additionally investigated the use of bioad-hesives to attach the electrodes to the retina but did not make any specificrecommendations.37 As these studies are ongoing, the safety and efficacy oftheir epiretinal approach is yet to be established; however, it is clear that thisgroup is addressing the key issues.

While performing this animal research, this group also studied the per-cepts resulting from electrical stimulation of the retina in blind human vol-unteers.38–40 Initially, they investigated the most basic question: Does electri-cal stimulation by bipolar electrodes, manually placed on the retinal surface,evoke a percept in the blind? In a sense, this experiment repeated their earlieranimal experiments but here they could ask the more germane question:what current level leads to a percept? They found that percepts could beevoked with charge densities of threshold between 160 to 3200 µC/cm2 withplatinum electrodes. These percepts were reported to range in size from apinhead to the size of pea, at a distance of 30 cm. Further, the visuotopicorganization of percepts, in a wide sense, was verified by stimulating widelyseparated regions of the retina. Interestingly, the size of the percept, and itsbrightness, did not appear to vary with retinal location. In a second set ofexperiments in blind volunteers, these researchers investigated the electrodespacing necessary to evoke distinct percepts. At the minimum spacing tested,435 µm, the two phosphenes were distinct and separated by just over 1degree. This spacing is consistent with the known magnification factor ofthe retina, approximately 250 µm per degree of visual space. In these exper-iments, the electrodes had diameters in the 50- to 100-µm range.

In a third set of experiments, using an array of electrodes, arranged eitheras a 5 × 5 grid or a 3 × 3 grid with the 400-µm-diameter platinum electrodesat 600 µm center-to-center separations, the percept of a fused line could beevoked by the stimulation of a line of electrodes on the array. Further,stimulating either a row or column of electrodes evoked distinct line per-cepts. When stimulating a U-shaped and box-shaped grouping of electrodes,the percepts were reported as an H shape and a box shape, respectively. Thereliability of these shaped percepts was not reported. Due to incomplete orinconsistent information in their published reports, it is difficult to estimatethe charge density necessary to evoke these percepts. It appears that thresh-old for the H-shaped percept required a charge density of about 300µC/cm2.40 This charge density is the theoretic maximal safe charge density41



but greatly exceeds the practical safe limit (25 to 75 µC/cm2) for long-termstimulation with platinum.17 These investigators recognize that their chargedensities would be unsafe in a long-term neuroprosthesis32 and have pro-posed overcoming this problems by either using activated iridium elec-trodes, with a practical safe charge density of up to 3000 µC/cm2, or usingdifferent electrode configurations.40 With 600-µm center-to-center spacing ofelectrodes, the resulting phosphene centers would be separated by 2.4degrees, a rather poor pixelization of the visual world. Yet, to increase theresolution, the electrode spacing would have to be reduced, resulting inreduced electrode size and increased current density.

11.5.2 MIT and Harvard group

The MIT and Harvard collaboration has primarily investigated the engineer-ing aspects of an epiretinal implant but more recently they have begun topursue human psychophysical experimentation as a proof of concept.32 Theirearlier work investigated the microfabrication and electronic design of thestimulating electrodes as well as the signal-processing electronics. This, ofcourse, led to studies of the stimulation parameters necessary to evokepercepts and power requirements of the necessary devices.42 Although theirhuman psychophysical results have only been presented in abstract andpreprint form, their results indicate that, while percepts can be induced byelectrical stimulation of the retina, the ability to generate a patterned perceptappears limited. The pattern of the percept did not follow the anticipatedpattern in the majority of the cases and the reliability of the percept (similardescription for the same pattern of stimulation) was only 66%.18 Interestingly,a normally sighted volunteer had an 82% reliability.44 It is unclear whetherthe normally sighted volunteer gave a more reliable report due to being moreaccustomed to visualizing objects or if this is an indication that the blind’sdysfunctional retina is more disorganized than expected. Further, the chargedensities necessary to produce a percept, up to 28 µC/cm,2 approached thesafe, long-term limit.44

Their animal research has centered on both the long-term ability to evokeactivity in the visual cortex, measured by evoked potentials, in rabbits andthe question of what neural structures are excited by electrical stimulationof isolated rabbit retina.45 This latter research has opened the possibility ofinvestigating stimulation paradigms with arrays of electrodes that effectivelyproduce patterned activity. In these experiments, a microfabricated electrodearray was used for both stimulation and recording. The platinum black-coated, gold-stimulating electrodes were 10 µm in diameter and were placedon 25-µm centers. The recording electrodes had the same diameter and wereplaced on 70-µm centers. The retina was stimulated with anodic first, bipha-sic current pulses of between 0.01 and 20 µA amplitude, lasting 400 µsec perhalf phase. The threshold for observing an extracellular action potentialranged from 0.06 to 1.8 µA. However, the method appeared to very prefer-entially excite the passing ganglion cell axons and not the underlying retinal



ganglion cells. Generally, the observed extracellular action potential was heldto be due the ganglion cells soma becoming active due to antidromic prop-agation. Clearly, it is more desirable to provide somatic stimulation, allowingone to take advantage of the visuotopic organization of the ganglion cellsbodies. However, it is not necessary that somatic stimulation be used.

11.5.3 Northern German group

The northern German consortium, centered in Bonn, has primarily investi-gated the signal-processing aspects of an epiretinal implant. To provide amore natural percept, this group started the development of specializedelectronics, called a retinal encoder, that mimic the function of the retina andcan automatically adjust its operation based upon conversational input fromthe user.15 More recently, they have started an active program of primateexperimentation but much of their results in this area are only available inabstract form. In these tests, they verified that their neural interface, calledthe Multi-Microcontact Array (MMA), could be secured to the retina byretinal tacks and that cortical activation could be had by stimulation.46 How-ever, these tests were of limited duration, lasting 6 to 8 hours, and theirprincipal purpose appeared to be defining the precision of locating the MMA.The functionality of the array was tested only in a very gross sense and,then, no details the stimulus parameters were provided. Only limited histo-logical examinations were performed.

These results have been expanded upon by optically imaging the acti-vation of visual cortex in the cat while stimulating patterns of electrodes onthe retina. The results indicate the existence of a relationship between thestimulation pattern and the pattern of activity in primary visual cortex.47

However, the occurrence of a similar pattern of activity in visual cortex doesnot speak to the associated percept.

The northern German group has also performed computation analysesin an attempt to resolve the most likely neural structure activated by stim-ulation. In what is likely the most extensive study of all three groups, thisgroup argues that one can selectively stimulate the desired target, the somaof retinal ganglion cells, by the simultaneous control of current at multipleelectrode sites.48

11.5.4 Summary

Despite its relatively late entry into vision neuroprosthesis research, theepiretinal approach has made great strides but there are many yet to beresolved problems associated with this approach. The ability to access theneural retina has been established with novel surgical approaches. A generalproof of concept, that electrical stimulation of the inner surface of the retinaevokes action potentials, has been shown with isolated animal retinas.33,45

The Humayun group has shown the more specific proof of concept thatpatterned visual percepts can be invoked by patterned electrical stimulation,



a result found in blind human volunteers with either AMD or RP.38–40 How-ever, in more controlled studies, the MIT group has found only a weakrelationship between a percept and the anticipated percept due to electricalstimulation. Blind volunteers reported a percept that had a “reasonablerelationship” to the anticipated percept only around 40% of the time. Furtherthe reliability of the percept was low, with the same precept being describedonly 66% of the time when presented with the same pattern of electricalstimulation.18 Using an alternative approach to investigate patterned stimu-lation, the northern German group has shown that stimulation of the innerretina by an array of electrodes results in a visuotopically organized patternof activity in primary visual cortex, as observed by intrinsic optical imaging.47

Although an excellent method of investigating the relationship betweenelectrical stimulation and cortical activation, this method says little aboutthe relationship between patterned stimulation and perception. Collectively,these results argue that patterned perception may be possible but that it isnot necessarily assured and further investigations are warranted. Well-defined and appropriately implemented psychophysical tests, particularlyones addressing the reliability of the percept, must be performed before thisquestion is fully resolved.

An associated issue concerns which neural structures are being activated.Given the organization of the retina, it is possible that the passing retinalganglion cell fibers, which lay between the stimulating electrodes and thesoma of the retinal ganglion cell, are being activated. A number of compu-tational studies that have addressed the issue34,48 have concluded that thecell bodies can be preferentially activated, but their results have not beenverified in a biological preparation. The one attempt to study preferentialactivation in an excised retina used a vastly different electrode configurationand did not provide any means of extrapolating to the configurations pro-posed by other groups.45 However, if this research were expanded upon, onecould readily investigate how to preferentially stimulate the cell bodies ofthe retinal ganglion cells.

To ensure long-term function and stability of perception, the array ofstimulating electrodes must be securely attached to the retina. As the retinais a thin, delicate structure, just the fixation of a passive device to the retinais problematic. Further, as the electrode array and signal processing areactive devices, there are additional concerns regarding the neural tissueresponse to both chronic electrical stimulation and heating, the later dueto the power dissipations by the electronics. Preliminary results are avail-able that indicate that both the electronics and electrodes can be securedto the retina via retinal tacks without serious complications.35 These resultsare just now being expanded upon to include the effect of chronic electricalstimulation on tissue.36

Although the key assumptions introduced earlier have only partiallybeen addressed, the three principal research groups are clearly investigatingthe crucial issues. It appears that retinal tacks can attach thin film electrodearrays to the retina without serious complications. To induce a neural



response, the charge density often exceeded the safe, long-term limit.17 Inreaction, most epiretinal devices utilize relatively large electrodes but thisleads to relatively widely spaced phosphenes. Finally, the reliability of apatterned percept is presently under study for the epiretinal approach.

11.6 Optic nerve neuroprosthetic approachThe optic nerves provide the sole conduits of visual information from theretina to the lateral geniculate nucleus in the thalamus. In certain visualpathologies, such as age related macular degeneration or retinitis pigmen-tosa, where a subpopulation of ganglion cells has been spared, but in whichcells of the outer retina are completely degenerate, the optic nerve offers apossible site for intervention via a neural interface.

A group of researchers from the Catholic University of Louvain in Brus-sels has seriously entertained this as a candidate intervention site for a visualprosthesis.49,50 They have tested the concept in a series of experiments con-ducted over the past 4 years in a single human volunteer with retinitispigmentosa. The team has implanted a self-sizing spiral cuff electrode arrayaround the right optic nerve of this volunteer. The electrode array consistsof four surface electrodes on the inner surface of the cuff. When implantedaround the optic nerve, currents can be passed in a bipolar fashion betweengroups of these surface electrodes in order to achieve some degree of stim-ulation selectivity in the population of nerve fibers. The use of cuff electrodesto stimulate peripheral nerves has a long history.51 In many cases, suchelectrodes can provide long-term stimulation, with little biological compli-cations.52 Recent improvements in the implanted system involve wirelesstelemetry of stimulus signals to the implanted array.

The surgical access used in the implantation is complex. The lead wiresin the implant system were fed under the skin, down the neck to a telemetryunit implanted near the collarbone. In this one subject, there has been noreport of complications from the surgery or breakage of the lead wires.

The subject has been stimulated intermittently over the 4-year period ofthe implantation, and studies of phosphene thresholds and their spatiallocation have been monitored. Wide ranges of phosphenes have been ableto be evoked with this system. They span a region extending 85 horizontaldegrees and 60 vertical degrees in front of the observer (the observer pointsto the perceived phosphene location). The phosphenes range in size from 1to 50 square degrees, and for a given bipolar stimulation regime the locationof the phosphene, its intensity, and its size are a function of the stimulationcurrent. As the stimulus current is increased by a factor of three, the locationof the perceived phosphene migrates from the edge of the phosphene spaceto the center of the space. The absolute thresholds for evoking a just percep-tible phosphene vary with stimulus pulse duration and whether the stimu-lation is delivered as a train or as a single biphasic pulse. For single biphasicpulses of 213 µsec durations, the average thresholds were about 350 µA,while for 17 pulse trains, delivered at 160 Hz, the thresholds for perceptions



dropped to about 15 µA. When the stimuli were delivered at a low repetitionrate, the phosphenes were observed to flicker, but they fused into a steadypercept when the stimulus rate was between 8 and 10 Hz. The steady perceptproduced by these stimuli fades out after 1 to 3 seconds. The brightness ofall evoked phosphenes varied with stimulus intensity, and ranged from“dim” to “average” as the current strength was increased by a factor of three.

The researchers claim that the phosphene space is sufficiently stable thatthey have been able to predict a phosphene’s size, location, and intensityfrom the stimulation parameters used to evoke the phosphene. This is clearlya critical issue if optic nerve stimulation is expected to restore any form ofuseful vision.

A useful visual sense can only be built from multiple phosphenes,evoked at predictable locations. The Belgium team has used interleavedstimulation to evoke patterns of from 4 to 24 phosphenes, and the subjecthas been able to identify simple objects using these phosphene fields byscanning the objects using a head-mounted camera.

It is clear from these observations that this approach is unlikely to beable to recreate a high-resolution visual sense in those implanted with thefour-electrode cuff. However, the researchers suggest that visually guidedmobility and some forms of task performance do not necessarily require ahigh-resolution visual sense. They suggest that with sufficient training, asubject implanted with an optic nerve array could use this limited visualinput to achieve simple visually guided mobility (at least in familiar envi-ronments). However, a number of issues require additional research beforethis approach could be considered tenable. Issues that relate to this potentialintervention site are (1) how significant and how viable is the population ofoptic nerve fibers that are still functional; (2) does electrical stimulation ofthe optic nerve spare it from continued degeneration or does constant stim-ulation accelerate the process; (3) how many phosphenes, evoked by opticnerve stimulation, are required to produce given levels of visual task per-formance; and (4) would other electrode array designs that penetrate intothe optic nerve produce more focal stimulation of optic nerve fibers andbetter control of phosphene location and intensity?

Researchers at the University of Utah have conducted experimentsaddressing this last issue. They have developed a unique electrode arrayarchitecture that was designed to provide highly selective electrical accessto a number of the nerve fibers in the sciatic nerve.53 This device, the UtahSlanted Electrode Array (USEA), is shown in Figure 11.7, and its access tothe fibers in the nerve is depicted in Figure 11.8. It is built from silicon andcontains 100 electrodes designed to penetrate the epineurium that surroundsthe nerve and the perineuria that surround the individual fascicles withinthe nerve. The length of each electrode varies along the length of the array,with 0.5-mm-long electrodes on one side of the array, and 1.5-mm-longelectrodes on the opposite side of the array. Each electrode is electricallyisolated from its neighboring electrodes with a moat of glass at its base, andeach electrode has a lead wire connected to a bond pad at its base that is



Figure 11.7 Potential neural interface, the USEA, for access to the optic nerve. The array of 100 electrodes, shown to the left, has beensuccessfully used in peripheral nerve of cats. As each row of 10 electrodes has a unique length, ranging from 0.5 to 1.5 mm, each rowof electrodes accesses a unique depth in the nerve, as shown in the center panel. Thus, this structure could potentially access 100unique axonal fibers within a nerve, as illustrated in the right panel. The distance between rows (and columns) of electrodes is 400 µm.



brought out to a percutaneous connector. The tips of each electrode aremetalized with platinum or iridium to facilitate the transduction of electronsin the wires into ions in the nerve. The entire structure, with the exceptionof the metalized tips, is insulated with silicon nitride or polymeric materials.

The Utah researchers have inserted the USEA into the sciatic nerve ofthe cat and measured the amount of electrical current required to evoke atwitch in the muscles innervated by the motor neurons of the sciatic nerve.53

They found that single biphasic current pulses in the 5- to 20-µA range weregenerally sufficient to evoke muscle twitches. They also studied selectivityof stimulation by monitoring which muscles were excited when currentswere passed through each of the implanted electrodes. They found that eachof the muscles of the lower leg and ankle could be individually stimulatedwith currents passed through appropriate electrodes in the implanted array,and that individual muscles often could be activated by multiple electrodesin the array. Finally, these electrodes appeared to activate specific indepen-dent subsets of motoneurons. Such an electrode array architecture holdspromise for an optic nerve interface.

11.7 Cortically based neuroprosthetic approachAs described earlier, the visuotopic organization of the visual pathways hasmotivated much of the effort at developing retinal, optic nerve, and corticallybased vision neuroprostheses. An artist’s conception of what a corticallybased visual prosthesis might look like is shown in Figure 11.8. While apractical cortically based visual prosthesis has yet to be developed, theconcept was first seriously entertained about 30 years ago. Before describing

Figure 11.8 Artist’s concept of a cortically based visual neuroprosthesis. A smallcamera, located in the bridge of a pair of glasses, encodes the visual scene. This signalthen drives a small array of electrodes implanted into primary visual cortex. Thesignal-processing electronics, not shown, convert the video signal into controlledcurrent pulses on each of the electrodes, resulting in the percept of the same patternseen by the camera.



recent work in this field, we will explore the historical foundations of thisapproach to limited sight restoration.

The earliest work in this field was focused on implanting electrodesarrays on the surface of the primary visual cortex; this work was motivatedby the pioneering findings of Hubel and Wiesel54 and others that the visualcortex also had a visuotopic organization. Further, while access to the pri-mary visual cortex requires specialized neurosurgical techniques, this regionof cerebral cortex is surgically accessible without great difficulty, and it isprotected by the skull from mechanical insult. Thus, the primary visualcortex seemed to be a good target for a first generation visual prosthesis.

11.7.1 Historical overview

Two efforts were launched in the 1960s and 1970s to explore this possibility:one by Brindley in England,55 and one by Dobelle and colleagues at theUniversity of Utah.56 Both approaches proposed to stimulate visual phos-phenes using arrays of platinum electrodes implanted under the dura, buton the surface of the visual cortex. The main difference between these twoapproaches was in the means by which stimulation signals were to be passedacross the scalp. Brindley had developed an array of inductively coupledtransmitting coils that were worn over the scalp and receiving coils attachedto each electrode that were implanted under the scalp. Dobelle used a per-cutaneous connector mounted behind the ear, with each electrode connectedto one of 64 pins in the connector.

Patients were implanted with these devices and extended periods oftesting were conducted. The subjects were able to perceive points of lightwhen currents were passed into the visual cortex, but two problems madethe approach impractical. First, the currents required to evoke visual phos-phenes were in the 1- to 10-mA range, current levels deemed unsafe for long-term chronic stimulation. Further, if future generation systems would requirelarger numbers of electrodes, summated currents could reach dangerouslevels that could produce seizures. Second, the currents passed throughelectrodes evoked phosphenes that interacted in a nonlinear fashion. Specif-ically, the location of phosphenes that were evoked with current injectionsthrough individual electrodes was altered when currents were passedthrough groups of electrodes. This meant that the electrodes would have tobe spaced at large intervals, making the concept of contiguous phosphenebased percepts impossible. However, the experiments provided additionalsupport for the visuotopic organization of human visual cortex and demon-strated that patterned electrical stimulation could evoked discriminatablepatterned percepts: When stimulated with a subset of six electrodes, subjectscould read Braille characters faster than with their sense of touch.57

The cortically based vision neuroprosthetic field then lay fallow for anumber of years until a group of researchers in the Neuroprosthesis Programat the National Institute of Health conducted a series of acute experimentsusing electrodes that penetrated into visual cortex of human volunteers.58,59



The experiments were based upon the premise that, because the normalinput to the visual cortex was layer 4C, electrodes that penetrated the visualcortex to this depth could produce much more focal stimulation of corticalneurons, with much lower current levels than were achieved with surfaceelectrode arrays. This hypothesis was borne out in their experiments whereit was shown that phosphenes could be evoked with electrical currents inthe 1- to 10-µA region and that two distinct phosphenes could be evoked bycurrent injections into electrodes as closely spaced as 500 µm. The electrodesused by the NIH team were of a map pin architecture. Each electrode wasat the end of a 1.5-mm-long insulated needle, and each had a thin insulatedwire bonded to the end of the needle. The interconnection of the wire andthe electrode was insulated with a small bead (creating the map pin shape).The NIH team also used pairs and triplets of map pin electrodes to investi-gate the dependence of evoked phosphenes on electrode spacing. Whilethese experiments demonstrated the benefits of penetrating electrodes oversurface electrodes, the use of individually inserted map pin electrodes wasan impractical approach to implantation of large numbers of electrodes.

11.7.2 Electrode arrays

The problem of creating high-count arrays of penetrating microelectrodesthat could be easily implanted into cerebral cortical tissues was attackedindependently by two research teams: one led by Wise at the University ofMichigan,60 and one led by Normann at the University of Utah.61 Both groupswanted to capitalize upon newly developed silicon microfabrication tech-nologies to build high-count arrays of penetrating microelectrodes becauseof the biocompatibility of silicon. Both teams are capable of making electrodearrays with hundreds of needles, and both teams have demonstrated thatthese arrays can be implanted chronically, often with little tissue insult. Thedifference between the arrays is in the way that they are fabricated. TheMichigan array is built using conventional photolithographic techniques thatresult in very precise and reproducible array geometries. The Utah ElectrodeArray is built using micromachining techniques and requires more hands-on effort. Each needle in the Utah array has an approximately cylindricalcross section that rends the cortical tissues it is inserted into rather thancutting them. The Michigan arrays can have multiple electrode sites on eachneedle, and each needle has a rectangular cross section and may cut ratherthan rend the cortical tissues it is inserted into.

11.7.3 Animal experiments

These complex electrode array geometries were designed to be implantedinto cerebral cortex to a depth of 1.5 to 2mm; however, the Utah teamdiscovered that arrays containing very large numbers of individual elec-trodes cannot simply be pushed into the cortex of experimental animals.Rather, just as a staple can be inserted into a block of wood if it has sufficient



momentum and velocity, these arrays require high velocity insertion to fullyinsert them. To achieve this end, the Utah team has built a pneumaticallyactuated insertion tool.62 This insertion tool is capable of full insertion of a100-electrode array in fewer than 200 µsec (an insertion velocity on the orderof 7 m/sec).

Both the Utah and the Michigan electrode arrays have been used inscores of electrophysiological experiments in a variety of sites in the nervoussystem in rats, turtles, cats, monkeys, guinea pigs, and ferrets. They havebeen used in both stimulation and recording applications in both acute andchronic preparations. Histological studies indicate that there is usually a littlelocalized bleeding associated with their implantation, but this usuallyresolves itself quickly, and single- and/or multi-unit recordings can be madewithin an hour or two of implantation. The fact that single-unit recordingscan be made with these complex devices provides the best evidence that theneurons near the electrode tips have not been significantly damaged by theinsertion process. The fact that recordings can be made on a chronic basisindicates that the materials used in their manufacture are biocompatible.

The efficacy of the Utah Electrode Array as a means to stimulate neuraltissues has been studied in a series of chronic behavioral experiments con-ducted in cats.63 Rather than training cats to respond to a visual task, Rouschedecided to train cats to respond to auditory stimuli (cats seem to alwaysattend to auditory stimuli while this is not the case for visual stimuli). Catswere trained to press a lever whenever they detected an auditory tonedelivered through a loudspeaker that was initiated by a different lever press.When proficient at this task, the cats were implanted in the auditory cortexwith a Utah array. After the implant, the cats were periodically electricallystimulated via the implanted array rather than aurally stimulated. If theelectrical stimulation evoked a percept, the cats pressed the lever. Theseexperiments demonstrated that the cats were able to detect presumed elec-trically evoked auditory percepts for current injections in the 1- to 10-µArange, a range well within the safe limits of chronic stimulation.

11.7.4 Human experimentation

The success of these animal experiments suggests that the stage may be setfor short-term human experimentation with both the Michigan and the Utahelectrode arrays. While such experimentation has yet to be conducted, issuesto be resolved are:

Can phosphenes be evoked in every electrode that is implanted in visualcortex?

What is the variation in threshold currents required to evoke suchphosphenes?

What is the nature of the phosphenes from electrode to electrode?Does stimulation of each electrode always evoke spots of light or do

some electrodes evoke spots of darkness?



What is the stability of the phosphene thresholds over time?What is the visuotopic organization of the stimulated electrodes and

the evoked phosphenes?What is the minimum separation between stimulated electrodes that

evokes two separate phosphenes?How does current injection in one electrode affect phosphene percepts

evoked in neighboring electrodes?Do simple spatial patterns of electrical stimulation evoke patterned

percepts that can be discriminated from other patterns of electricalstimulation?

The answers to these questions will provide the most complete proof ofconcept to date that electrical stimulation of visual cortex via an array ofpenetrating electrodes could eventually restore a useful visual sense in indi-viduals with profound blindness. This will set the stage for a series of chronichuman experiments where electrodes will be stimulated with more complexspatial patterns, and eventually with signals originating from a portablevideo camera.

11.8 ConclusionsThis chapter has provided a brief overview of the physiological foundationsupon which visual neuroprosthetic systems have been built. The chapter hasdescribed four intervention sites where a neural interface could be effectivelyused to stimulate neurons in the visual pathways: the subretinal sites, epiret-inal sites, the optic nerve, and the primary visual cortex. Research in thefield of visual neuroprostheses has been invigorated over this past decadedue to recent advances in microfabrication of physical devices, VLSI elec-tronic circuitry, and wireless telemetry. These technological innovations, cou-pled with improved understanding in the systems-level sensory neuro-science, are making the field of neuroprostheses into a reality. The authorsof this chapter hope that the readers understand and appreciate four mainpoints that were elaborated upon in the chapter:

• The use of small, passive photodiode arrays cannot produce sufficientcurrents to focally excite second-order retinal neurons with physio-logical levels of retinal illumination. Any such visual neuroprosthesiswill require active electronics to boost the relatively small signals thatare produced by small photodiodes.

• In order to optimally excite neurons, it will be important to positionthe stimulating electrodes very close to the neurons one is trying toexcite (within a few microns). This means that the electrode must havedimensions that are similar to the size of the neurons one is trying tostimulate. Because of the location of the neurons within tissues, thisconstraint argues for a penetrating (or needle-shaped) electrode geom-etry and argues against the use of planar or surface electrodes.



• Stimulation of neurons is best accomplished with highly localizedcurrent injections. This again is best achieved with a penetratingelectrode geometry.

• The recent work in visual neuroprosthetic systems has not yet result-ed in the development of functional clinical systems. However, thiswork has resulted in the development of new classes of basic researchinstruments (electrode arrays) that will enable researchers to betterunderstand the physiology of neuronal stimulation and sensory sig-nal processing. These devices can be used in animals, and are nowbeginning to be used in human volunteers.

It is clear from what we have presented in this chapter that the field ofvisual neuroprostheses is rapidly emerging and that clinical systems arelikely to become available over the next decade. What is also clear is thatresearch should be continued in developing systems that can be appliedat the retinal, optic nerve and cortical levels. Successful visual neuropros-thetic systems in all three interventional sites will give the ophthalmicsurgeon or the neurosurgeon entirely new approaches that will providelimited but functional restoration of sight in those who have lost this highlyprized sensory modality.

References1. Nicholls, J.G., Martin, A.R. et al., From Neuron to Brain: A Cellular and Molecular

Approach to the Function of the Nervous System, Sinauer Associates, Sunderland,MA, 1992.

2. Kandel, E.R., Schwartz, J.H. et al., Principles of Neural Science, McGraw-Hill,New York, 2000.

3. Balazsi, A.G., Rootman, J. et al., The effect of age on the nerve fiber populationof the human optic nerve, Am. J. Ophthalmol., 97(6), 760–766, 1984.

4. Naito, J., Retinogeniculate projection fibers in the monkey optic chiasm: ademonstration of the fiber arrangement by means of wheat germ agglutininconjugated to horseradish peroxidase, J. Comp. Neurol., 346(4), 559–571, 1994.

5. Fitzgibbon, T. and Taylor, S.F., Retinotopy of the human retinal nerve fibrelayer and optic nerve head, J. Comp. Neurol., 375(2), 238–251, 1996.

6. Hubener, M., Shoham, D. et al., Spatial relationships among three columnarsystems in cat area 17, J. Neurosci., 17(23), 9270–9284, 1997.

7. Van Essen, D.C., Anderson, C.H. et al., Information processing in the primatevisual system, an integrated systems perspective, Science, 255(5043), 419–423,1992.

8. Merigan, W.H. and Maunsell, J.H., How parallel are the primate visual path-ways?, Annu. Rev. Neurosci., 16, 369–402, 1993.

9. Kobatake, E. and Tanaka, K., Neuronal selectivities to complex object featuresin the ventral visual pathway of the macaque cerebral cortex, J. Neurophysiol.,71(3), 856–867, 1994.

10. Stone, J.L., Barlow, W.E. et al., Morphometric analysis of macular photorecep-tors and ganglion cells in retinas with retinitis pigmentosa, Arch. Ophthalmol.,110(11), 1634–1639, 1992.



11. Santos, A., Humayun, M.S. et al., Preservation of the inner retina in retinitispigmentosa: a morphometric analysis, Arch. Ophthalmol., 115(4), 511–515,1997.

12. Kim, S.Y., Sadda, S. et al., Morphometric Analysis of the Macular Retina from Eyeswith Discform Age Related Macular Degeneration, ARVO, Fort Lauderdale, FL,2001.

13. Jones, B.W., Chen, C.K. et al., Severe Remodeling of the Mouse Neural RetinaTriggered by Rod Degeneration, ARVO, Fort Lauderdale, FL, 2002.

14. Warren, D.J., Fernandez, E. et al., High-resolution two-dimensional spatialmapping of cat striate cortex using a 100-microelectrode array, Neuroscience,105(1), 19–31, 2001.

15. Eckmiller, R., Learning retina implants with epiretinal contacts, OphthalmicRes., 29(5), 281–289, 1997.

16. Jones, K.E. and Normann, R.A., An advanced demultiplexing system forphysiological stimulation, IEEE Trans. Biomed. Eng., 44(12), 1210–1220, 1997.

17. Robblee, L.S. and Rose, T.L., Electricochemical guidelines for selection ofprotocols and electrode materials for neural stimulation, in Neural Prostheses:Fundimental Studies, A.W.F. and McCreery, D.B., Eds., Prentice-Hall, Engle-wood Cliffs, NJ, 1990, pp. 25–66.

18. Rizzo, J.F., III, Wyatt, J. et al., Retinal prosthesis, an encouraging first decadewith major challenges ahead, Ophthalmology, 108(1), 13–14, 2001.

19. Chow, A.Y. and Chow, V.Y., Subretinal electrical stimulation of the rabbitretina, Neurosci. Lett., 225(1), 13–16, 1997.

20. Chow, A.Y. and Peachey, N.S., The subretinal microphotodiode array retinalprosthesis, Ophthalmic Res., 30(3), 195–198, 1998.

21. Peyman, G., Chow, A.Y. et al., Subretinal semiconductor microphotodiodearray, Ophthalmic Surg. Lasers, 29(3), 234–241, 1998.

22. Chow, A.Y. and Peachey, N., The subretinal microphotodiode array retinalprosthesis, II, Ophthalmic Res., 31(3), 246, 1999.

23. Peachey, N.S. and Chow, A.Y., Subretinal implantation of semiconductor-based photodiodes, progress and challenges, J. Rehabil. Res. Dev., 36(4),371–376, 1999.

24. Chow, A.Y., Pardue, M.T. et al., Implantation of silicon chip microphotodiodearrays into the cat subretinal space, IEEE Trans. Neural Syst. Rehabil. Eng., 9(1),86–95, 2001.

25. Pardue, M.T., Stubbs, Jr., E.B. et al., Immunohistochemical studies of theretina following long-term implantation with subretinal microphotodiodearrays, Exp. Eye Res., 73(3), 333–343, 2001.

26. Chow, A.Y., Peyman, G.A. et al., Safety, Feasibility and Efficacy of SubretinalArtificial Silicon Retina™ Prosthesis for the Treatment of Patients with RetinitisPigmentosa, ARVO, Fort Lauderdale, FL, 2002.

27. Zrenner, E., Miliczek, K.D. et al., The development of subretinal microphoto-diodes for replacement of degenerated photoreceptors, Ophthalmic Res., 29(5),269–280, 1997.

28. Zrenner, E., Stett, A. et al., Can subretinal microphotodiodes successfullyreplace degenerated photoreceptors?, Vision Res., 39(15), 2555–2567, 1999.

29. Zrenner, E., Will retinal implants restore vision?, Science, 295(5557),1022–1025, 2002.

30. Guenther, E., Troger, B. et al., Long-term survival of retinal cell cultures onretinal implant materials, Vision Res., 39(24), 3988–3994, 1999.



31. Stett, A., Barth, W. et al., Electrical multisite stimulation of the isolated chickenretina, Vision Res., 40(13), 1785–1795, 2000.

32. Rizzo, J.F., Wyatt, J.L. et al., Accuracy and Reproducibility of Percepts Elicited byElectrical Stimululation of the Retinas of Blind and Normal Subjects, ARVO, FortLauderdale, FL, 2001.

33. Humayun, M., Propst, R. et al., Bipolar surface electrical stimulation of thevertebrate retina, Arch. Ophthalmol., 112(1), 110–116, 1994.

34. Greenberg, R.J., Velte, T.J. et al., A computational model of electrical stimula-tion of the retinal ganglion cell, IEEE Trans. Biomed. Eng., 46(5), 505–514, 1999.

35. Majji, A.B., Humayun, M.S. et al., Long-term histological and electrophysio-logical results of an inactive epiretinal electrode array implantation in dogs,Invest. Ophthalmol. Vis. Sci., 40(9), 2073–2081, 1999.

36. Weiland, D., Fujii, G.Y. et al., Chronic Electrical Stimulation of the Canine Retina,ARVO, Fort Lauderdale, FL, 2002.

37. Margalit, E., Fujii, G.Y. et al., Bioadhesives for intraocular use, Retina, 20(5),469–77, 2000.

38. Humayun, M.S., de Juan, Jr., E. et al., Visual perception elicited by electricalstimulation of retina in blind humans, Arch. Ophthalmol., 114(1), 40–46, 1996.

39. Humayun, M.S. and de Juan, Jr., E., Artificial vision, Eye, 12(pt. 3b), 605–607,1998.

40. Humayun, M.S., de Juan, Jr., E. et al., Pattern electrical stimulation of thehuman retina, Vision Res., 39(15), 2569–2576, 1999.

41. Brummer, S.B., Robblee, L.S. et al., Criteria for selecting electrodes for elec-trical stimulation, theoretical and practical considerations, Ann. N.Y. Acad.Sci., 405, 159–71, 1983.

42. Wyatt, J.L., Jr. and Rizzo, III, J.F., Ocular implants for the blind, IEEE Spectum,33, 47–53, 1996.

44. Rizzo, J.F., Personal communication, 2002.45. Grumet, A.E., Wyatt, Jr., J.L. et al., Multi-electrode stimulation and recording

in the isolated retina, J. Neurosci. Meth., 101(1), 31–42, 2000.46. Gerding, H., Hornig, R. et al., Implantation, Mechanical Fixation, and Functional

Testing of Epiretinal Multi-Microcontact Arrays, MMA in Primates, ARVO, FortLauderdale, FL, 2001.

47. Eysel, U.T., Walter, P. et al., Optical Imaging Reveals Two-Dimensional Patternsof Cortical Activation after Local Retinal Stimulation with Sub- and Epiretinal VisualProstheses, ARVO, Fort Lauderdale, FL, 2002.

48. Hornig, R. and Eckmiller, R., Movement of Stimulation Focus by Multi-ElectrodeClusterstimulation for Retina Implants: Computational Results, ARVO, Fort Lau-derdale, FL, 2001.

49. Veraart, C., Raftopoulos, C. et al., Visual sensations produced by optic nervestimulation using an implanted self-sizing spiral cuff electrode, Brain Res.,813(1), 181–186, 1998.

50. Veraart, C., Raftopoulos, D. et al., Optic Nerve Electrical Stimulation in a RetinitisPigmentosa Blind Volunteer, Society for Neuroscience, Los Angeles, CA, 1998.

51. Mortimer, J.T., Agnew, W.F. et al., Perspectives on new electrode technologyfor stimulating peripheral nerves with implantable motor prostheses, IEEETrans. Neural Syst. Rehabil. Eng., 39(15), 145–153, 1995.

52. Romero, E., Denef, J.F. et al., Neural morphological effects of long-term im-plantation of the self-sizing spiral cuff nerve electrode, Med. Biol. Eng. Comput.,39(1), 90–100, 2001.



53. Branner, A., Stein, R.B. et al., Selective stimulation of cat sciatic nerve usingan array of varying-length microelectrodes, J. Neurophysiol., 85(4), 1585–1594,2001.

54. Hubel, D.H. and Wiesel, T.N., Receptive fields, binocular interaction andfunctional architecture in the cat’s visual cortex, J. Physiol., 160, 106–154, 1962.

55. Brindley, G.S. and Lewin, W.S., The visual sensations produced by electricalstimulation of the medial occipital cortex, J. Physiol. (London), 194(2), 54–59,1968.

56. Dobelle, W. and Mladejovsky, M., Phosphenes produced by electrical stimu-lation of human occipital cortex, and their application to the development ofa prosthesis for the blind, J. Physiol. (London), 243(2), 553–576, 1974.

57. Dobelle, W.H., Artificial vision for the blind: the summit may be closer thanyou think, ASAIO J., 40(4), 919–922, 1994.

58. Bak, M., Girvin, J.P. et al., Visual sensations produced by intracortical micro-stimulation of the human occipital cortex, Med. Biol. Eng. Comput., 28(3),257–259, 1990.

59. Schmidt, E.M., Bak, M.J. et al., Feasibility of a visual prosthesis for the blindbased on intracortical microstimulation of the visual cortex, Brain, 119,507–522, 1996.

60. Hoogerwerf, A.C. and Wise, K.D., A three-dimensional microelectrode arrayfor chronic neural recording, IEEE Trans. Biomed. Eng., 41(12), 1136–1146, 1994.

61. Jones, K.E., Campbell, P.K. et al., A glass/silicon composite intracortical elec-trode array, Ann. Biomed. Eng., 20(4), 423–437, 1992.

62. Rousche, P.J. and Normann, R.A., A method for pneumatically inserting anarray of penetrating electrodes into cortical tissue, Ann. Biomed. Eng., 20(4),413–22, 1992.

63. Rousche, P.J. andNormann, R.A., Chronic intracortical microstimulation (IC-MS) of cat sensory cortex using the Utah Intracortical Electrode Array, IEEETrans. Rehabil. Eng., 7(1), 56–68, 1999.



chapter twelve

Motor prostheses

Richard T. Lauer and P. Hunter Peckham

Contents

12.1 Introduction12.2 Clinical applications of motor prostheses

12.2.1 Injury to the spinal cord12.2.2 Injury to the brain12.2.3 Diseases that affect neural function

12.3 Motor prosthesis design12.3.1 Stimulus delivery system12.3.2 Control unit12.3.3 Command interface

12.4 The first motor prostheses12.4.1 The cardiac pacemaker12.4.2 Hand and foot stimulators

12.5 Limitations of a motor prosthesis12.6 Commercially available motor prostheses

12.6.1 Upper extremity motor prostheses12.6.2 Lower extremity motor prostheses12.6.3 Organ system prostheses

12.7 Current research motor prosthetic systems12.7.1 Upper extremity motor prostheses12.7.2 Lower extremity motor prostheses12.7.3 Organ system prostheses

12.8 Future avenues of investigation12.8.1 New clinical applications12.8.2 Technology development

12.9 Summary and conclusionReferences



12.1 IntroductionDamage or disease of the central nervous system results in a variety ofdeficits including sensory loss, tonic contraction of muscle (spasticity), cog-nitive impairment, impairment of biological functions, and the loss of voli-tional control over the extremities. Current treatment methodologies forthese deficits include the use of pharmacological agents, physical therapyand rehabilitation, and surgical intervention. Also, other treatment method-ologies could be available at some time in the future, such as neural regen-eration. One method that is currently available and often overlooked for thetreatment of deficits resulting from central nervous system trauma, however,is the use of neural prostheses that effect motor function.

A neural prosthesis is a device that uses electrical stimulation to interfacedirectly to the nervous system to restore function. A motor prosthesis isdefined as a device that electrically stimulates a nerve (or nerves) innervatinga muscle (or series of muscles) for restoring functional movement or biolog-ical function. The purpose of this chapter, therefore, will be on how theprinciples of electrical stimulation, as realized in a motor prosthesis, can beused to overcome motor and functional losses. To this end, this chapter willbe structured to provide the reader with the following information:

• Clinical applications of the motor prosthesis• Characteristics/design of the motor prosthesis• Commercial and research motor prostheses• Future avenues of development for motor prostheses

12.2 Clinical applications of motor prosthesesThe motor prosthesis, as stated in the introduction, can be used to restoreloss of movement or biological function after damage to the central nervoussystem. Damage to the central nervous system can take several forms, theresult of either trauma or disease, and the effects of the damage can varygreatly. In fact, even with the same trauma or disease the effects on the centralnervous system can vary greatly from individual to individual. This presentsa unique problem for the use of motor prostheses as a rehabilitation tool.Therefore, before discussing motor prostheses and ongoing research in thisarea, it is first necessary to examine some of the diseases and traumas thatcan affect the central nervous system and define what the general needs ofthe individual will be given the pathology.

12.2.1 Injury to the spinal cord

The first area of central nervous system trauma to be discussed will beinjuries sustained to the spinal cord. The reasons for presenting this topicfirst are that the greatest levels of success with applications of motor pros-theses have been achieved with spinal cord injuries. Injuries of the spinal



cord result in different types of disorders depending upon the level of theinjury and the spinal cord tracts involved (Figure 12.1). Trauma to the spinalcord is usually the result of an automobile or diving accident, or a missilewound,1 often secondary to a fracture or dislocation of the vertebral column.This results in severance or compression of the cord by the fractured boneand edema (tissue swelling). As would be expected, the most mobile areasof the vertebral column are the most susceptible to injury. An injury to thelumbar or thoracic area results in paraplegia (paralysis of the lower extrem-ities), while an injury to the cervical area results in tetraplegia (paralysis ofboth the upper and lower extremities). Secondary to the loss of motor func-tion below the level of the injury is the subsequent death of the spinal nervesfrom those segments at the level of injury (denervation), muscle spasticity,and muscle atrophy. Muscle atrophy is the degeneration of the muscle tissuewith a resultant loss in muscle strength due to the loss of neuronal input.Because of this, the muscles that are to be stimulated by the motor prosthesiswill undergo a strengthening regimen to build up mass and strength beforethe introduction of the system.2

The loss of voluntary movement and the resultant effects upon motortasks are only one effect of spinal cord injury; there is also a subsequent lossin the autonomic functions of the nervous system. In the cases of highcervical level injuries, sustained at the second cervical level and higher, lossof the ability of the brain stem to control the diaphragm results in the inabilityto breathe voluntarily. In the case of both cervical and lumbar level lesions,bladder, bowel, and sexual functions are lost. The loss of these functions canlead to secondary complications, including urinary tract infections, impactedbowel and constipation, and difficulties with procreation.3

Figure 12.1 Overview of the human spinal cord and the motor/biological functionassociated with each level.



12.2.2 Injury to the brain

Injury to the brain can result in many different types of disorders dependingupon the areas involved (Figure 12.2). Most often, in addition to motordeficits, there are also deficits in cognitive and sensory functions. The cog-nitive and sensory effects of brain damage are outside of the scope of dis-cussion for the applications of a motor prosthesis. However, it is importantto note that the use of motor prostheses as a rehabilitation tool does dependupon the cognitive capabilities of the individual. If cognitive function isimpaired greatly by the extent of the injury to the brain, the effects of motorprostheses for the restoration of function can be hindered. However, motorprostheses can still be effective as tools for motor relearning. This will bediscussed later in this chapter.

One common cause of damage to the brain is that resulting from a cere-brovascular accident (CVA) or stroke. A stroke occurs when a blood vesselwithin the brain is either blocked or ruptured, resulting in a loss of blood flowto the area of the brain supplied by the vessel.1 This loss of blood flow leadsto neuronal death and subsequent losses in motor, sensory, and cognitivefunction depending on the size and location of the lesion. These losses, how-ever, may not be permanent and it is possible for individuals to learn tocompensate and correct for these deficits with rehabilitation therapy.4 The mostcommon form of stroke involves the motor area of the brain in one hemi-sphere.5 This results in hemiplegia (paralysis of the side of the body oppositeto the hemisphere involved). Secondary to the loss of motor function arecognitive changes, depending on which hemisphere is involved; hyperreflexia(exaggerated stretch reflexes); muscle spasticity (tonic levels of muscle con-traction independent of voluntary control); and muscle atrophy.

Another common form of brain injury is cerebral palsy. Cerebral palsyis the term used to define a wide range of movement disorders that occurduring or shortly after birth. These disorders are caused by an accident (fallsor other sudden trauma to the brain), brain infection (bacterial meningitisor viral encephalitis), or medical incidents during childbirth (such as birthasphyxia trauma, when oxygen flow to the brain has been cut off tempo-rarily).1,6 All of these conditions lead to neuronal death to parts of the brainsimilar to that experienced with stroke. Unlike stroke, however, the motoreffects are quite different. Most often (80% of the time), these children expe-rience what is known as spastic cerebral palsy.7 In spastic cerebral palsy, themuscles are stiffly and permanently contracted, leading to difficulties inperforming motor tasks. Other types of cerebral palsy include dyskineticcerebral palsy, characterized by uncontrolled, slow, writhing movements;ataxic cerebral palsy, which effects balance and depth perception; or a mixedform that combines the other three types to some degree.7

Brain injuries are also sustained by a quick and sudden trauma to thehead as can occur during an automobile or sporting accident. In this case,the trauma to the head results in the rupture of blood vessels on the surfaceof the brain and tissue swelling, causing an increase in intercranial pressure.



Figure 12.2 Schematic of cerebellum with approximate locations of important motor, sensory, and cognitive areas.




If not relieved, this increase in pressure results in neuronal death, and sub-sequent loss and impairment of motor function if the motor areas of thebrain are involved.

12.2.3 Diseases that affect neural function

Numerous diseases of the nervous system can result in a loss of neuromus-cular function. However, these can be roughly divided into those that leadto neuronal death, those that result in the loss of the myelin sheath aroundthe neuron preventing the conduction of action potentials, and those thataffect the generation or release of neurotransmitters. Of the three categories,for the application of motor prostheses only those diseases that cause neu-ronal death can be treated. More specifically, only those diseases that do notaffect the nerve going to the muscle have benefited from the use of thesesystems. Current motor prosthesis technology is dependent upon the artifi-cial generation of action potentials in the nerve going to the muscle to causemuscle contraction. Anything that affects action potential conduction in thenerve and the subsequent excitation of the muscle via the neuromuscularjunction will prevent effective use of the motor prosthesis. Therefore, theapplications of motor prostheses to the restoration of function in such dis-eases as multiple sclerosis and neuropathy are limited at this time.

Table 12.1 Summary of Neurological Conditions and the Associated Deficits

Note: Highlighted in the table are specific applications to which motor prostheses have beendeveloped and employed. Also indicated are what are believed to be possible futureavenues of motor prosthesis investigation.

Cerebral Palsy

Tetraplegia

Paraplegia

Multiple Sclerosis

Neuropathy

Amyotrophic Lateral Sclerosis

Parkinsonís Disease

Stroke

Mot

or D

efic

itsSp

astic

ityN

erve

Deg

ener

atio

nM

uscl

e At

roph

yC

ogni

tive

Def

icits

Sens

ory

Def

icits

Res

pira

tory

Def

icits

Blad

der/B

owel

Def

icits

x

Motor Prostheses Developed

Possible Future Research

x

O

x

x

x

x

x x

xx

O

O

O

O

O

O

O

O O O O O O

OOOOOO

O O O

O O

O O



Table 12.1 provides a summary of the expected deficits associated withbrain injury, spinal cord injury, and certain diseases of the nervous system.Not all the deficits listed will occur with each injury or disease, and eachindividual will present a unique case. The highlighted areas in the table arethose areas to which the motor prosthesis can and has been applied for useas a rehabilitation tool. Also given in the table are some suggestions of whatare believed to be possible future applications for motor prostheses. As canbe seen in this table, the motor prosthesis can be a valuable clinical tool forthe restoration of motor function, the restoration of biological functions, andfor motor relearning and training.

12.3 Motor prosthesis designThe applications of motor prostheses for the restoration of function haveproduced a variety of system designs, each directed toward a particularclinical use. However, all of the systems developed have certain character-istics in common and adhere to certain design criteria that are aimed atproviding the greatest functional benefit and user acceptance. The majorcomponents of the motor prosthesis system are:

• The stimulus delivery system, consisting of the electrode and leadwires, that provide stimulation of the nerve

• The control unit, responsible for interpreting the operational com-mand generated by the user and converting that information intomuscle stimulation

• The command interface, which records signals generated by the userand converts that information into operational commands for themotor prosthesis (Figure 12.3)

Figure 12.3 Block diagram showing the fundamental components of the motor pros-thesis and the interaction patterns with the intended user.

Senses

CommandInterface

CommandProcessor

ControlProcessor

Electrode

Environmental Feedback

Performance Feedback

Individual Control Unit Stimulus Delivery

MuscleActivation

Controlsignal



12.3.1 Stimulus delivery system

Currently, all motor prostheses operate by the electrical activation of nervesleading to muscles to elicit a muscle contraction. However, the methods bywhich the nerve can be stimulated are varied. One method by which a musclecan be electrically excited is with surface electrodes (Figure 12.4). Theseelectrodes are placed on the surface of the skin directly over the point wherethe nerve and muscle are joined (motor point). These electrodes require theuse of a conductive media and adhesives to ensure contact with the skin or,more commonly, are imbedded in an adhesive pad to ensure quick and cleanplacement.8,9 The advantages to the use of the surface electrode are therelatively low cost of the electrode and the noninvasive method of placement.The disadvantages to the use of these electrodes are that only those musclesfor which the motor point can be accessed from the skin can be electricallystimulated, and the placement of these electrodes requires an individual withknowledge and experience on the location of the motor point. In addition,the power requirements for such a system are usually higher given the largevoltages (approximately 80 V) required in order to drive the current acrossthe skin impedance.10

The second method by which the muscle can be electrically stimulatedis to use percutaneous electrodes (Figure 12.4). The percutaneous electrodeis inserted through the skin with a needle near the motor point of themuscle. The electrode can consist of a single strand of stainless steel wireor multiple strands.11,12 These electrodes are barbed at the end, usually bybending the end of the wire, to ensure that the electrode is anchored afterinsertion and is not removed with the needle. By crossing the skin barrier,the use of the percutaneous electrode reduces the amount of power

Figure 12.4 Schematic drawing representing the different types of electrode designs.Not represented in this drawing is the intramuscular (implanted) electrode type.

Surface

Percutaneous

Implanted

Epimysial Nerve Cuff



required to electrically stimulate the muscle and allows easier access tomuscles that lie deeper under the skin. The drawback to this method,however, is that the skin has been compromised, which can lead to skinirritation and possible infection. This requires greater diligence on the partof the motor prosthesis user, or, more likely, the attendant or caregiver, tokeep the entrance sites for the electrode clean. Percutaneous electrodes arealso susceptible to breakage due to stress at the skin–muscle interface,requiring the reinsertion of new electrodes.

A final method by which the muscle can be electrically stimulated iswith electrodes that are implanted completely within the body and do notcross the skin barrier (Figure 12.4). These are referred to as implanted elec-trodes, of which there are three major categories. The first type is the epimy-sial electrode, which consists of a platinum–iridium disk imbedded in areinforced silicon pad and sutured near the motor point of the muscle.13 Thesecond type of implanted electrode is the intramuscular electrode, which issimilar to the percutaneous electrode in that it is inserted with a needledirectly into the motor point of the muscle. The electrode is constructed ofstainless steel wire, but it has an umbrella-type of anchor at the end con-structed of polypropylene to maintain its position in the muscle.14 The finaltype of electrode is the nerve cuff electrode, which consists of one to threemetallic (generally platinum–iridium) bands imbedded in a silicon sheet thatencircle the nerve going to the muscle.15–17

The advantages to the use of the implanted electrodes are that in mostcases they are selective in the muscle that they stimulate, being located nearthe motor point of the muscle. The exception to this is the nerve cuff elec-trode. The use of this electrode can ensure selectivity only if the nervecontains motor nerves that innervate a single muscle. A nerve containingfibers for multiple muscles would have all of the muscles excited by the useof this electrode, unless other techniques were employed to ensure somedegree of selectivity.18 Other advantages to the use of the implanted elec-trodes are that they require less power to elicit a contraction within themuscle, which is of prime consideration in the development of implantedmotor prostheses. However, the primary drawback to their use is the levelof invasiveness, as surgery and multiple incision sites are required for theirplacement and replacement is difficult.

12.3.2 Control unit

The control unit of the motor prosthesis is the key component to the system,and consists of two parts (Figure 12.3). The first part is the command pro-cessor. The command processor is used to interpret user-generated signals,as recorded from a variety of human–machine interfaces, as command sig-nals to operate the various functions of the motor prosthesis. These functionscan include control over the system state (on/off/idle/lock) and selectionof activation patterns to be delivered to the muscles via the stimulus deliverysystem. The output of the command processor is relayed to the control



processor of the unit. The control processor converts the signals generatedby the command processor into actual function. The processor determines,based upon the activation pattern identified by the control processor, thespecific muscle stimulation levels necessary to achieve the pattern. It is thisinformation which is then delivered by the lead wires and electrodes to themuscle to generate purposeful contraction.

12.3.3 Command interface

The command interface, or human–machine interface, is the sensing systemthat records user-generated signals to operate the motor prosthesis (Fig-ure 12.3). Numerous types of user-generated signals have been used in thepast for control of motor prostheses. These include the use of muscle activity(electromyogram, or EMG) from a muscle under voluntary control,19–22 jointangle transducers,23,24 switches,25,26 respiration,27 voice activation,28–30 andeven a cortical signal.31 Independent of the type of control signal used, theinterface design has followed a set of criteria aimed at achieving optimalperformance and user acceptance. Several researchers have outlined thesecriteria as they pertain to both prosthetic and motor prosthetic control.23,32,33

What follows is a summary of these criteria, with subdivisions between theengineering aspects of the device and the user concerns.

The signal criteria set goals and expectations for the user-generatedsignal, allowing optimal performance of a motor prosthesis. The first crite-rion in this category for the command interface is to use the least numberof input channels possible to control all functions of the system. This isbecause studies have shown that as the number of inputs increase, controland accuracy can be negatively affected.34 The second criterion is to providethe greatest amount of information possible per input channel per unit oftime. This is referred to as information content and is expressed mathemat-ically as:35,36

I = T/

τ log2 n (12.1)

where I is information content in bits, T is transmission time in seconds

, τis the minimum time required for a level change in seconds, and n is thenumber of discrete levels.

Some studies have also defined the information outflow rate (IOR),related to the information content.37 The IOR considers the accuracy inachieving a discrete level and the transition rate. The equation is as follows:

IOR = (accuracy/transition rate)

× information content (12.2)

Information content is the same as calculated in Eq. (12.1).Another criterion of the signal is to reduce the amount of noise in the

signal, as measured by the signal-to-noise ratio (SNR). A value of 20 decibelshas been given as a possible acceptable value for motor prosthetic control.38



The final signal criterion refers to how quickly the input signal can changestates. This is defined as the transition rate for the signal.

The next category of criteria for an acceptable command interfaceincludes those that define the performance goals and the structure for thecontrol interface. The first is accessibility, which is defined as having directaccess to as many functions as possible. This implies the ability to multitask(parallel processing) instead of using menus and submenus to controlfunctions (serial processing). The next criterion is durability. The interfaceshould be able to withstand normal usage, and its lifetime should approachthat of the user. Interference is the concept that the interface should notinterfere or hinder the performance of normal activities or movement, norshould these activities interfere with the recording and interpretation ofthe signal. Processing is the measure of the interval between the initializa-tion of a command and the appropriate response of the system, involvingthe transition rate and signal processing involved. A maximum value forthis is approximately 200 msec, at which point the individual will beginto perceive the delay between thought and action.39 The design of theinterface should also be robust. This implies that the interface will providethe same level of performance without compensation by the user, regard-less of changes in placement or in the quality of the signal. In addition,the interface should be repeatable, responding with the same output for agiven input each time, with little drift or change. Finally, the interfaceshould have good resolution, allowing the user to generate both large andsmall changes in position and/or force.

The final category of criteria are those that reflect the wants and needsof the user as determined by socioeconomic factors. Cosmesis is impor-tant in that the controller should not draw additional attention to thedisability of the individual. The interface should also be as inexpensiveas possible and should be easy to put on and take off. Ideally, the usershould be able to accomplish this alone. Anything that is too complex orrequires a lengthy donning period could lead to rejection. Related to thisis the concept of ease of use, which implies that use in the controllerrequires a minimum training period and little concentration for use. Onemethod of achieving this is to use the concept of extended physiologicalproprioception (EPP).40,41 EPP is a prosthesis control technique that allowsthe user to accurately perceive the static and dynamic characteristics ofthe device through natural proprioceptive sensations. Finally, the inter-face should be free from all electrical hazards and the materials usedshould be biocompatable.

The objective with the design of the command interface to the motorprosthesis is to adhere to these criteria as closely as possible. Not all appli-cations of the motor prosthesis will require the use of all of these criteria,nor has there been an interface developed which has met all of these criteria.However, how close an interface comes to meeting these criteria ensureshow well the interface, and the motor prosthesis, will be accepted and usedby the potential recipient.



12.4 The first motor prosthesesThe previous sections of this chapter have provided the reader with anintroduction to the motor prosthesis by defining clinical applications of thedevice and outlining motor prosthesis design. At this point, it would nowbe worthwhile to review the predecessors of the modern-day motor pros-theses which were essential in defining the design criteria and providedinsight into how electrical stimulus could be used in the treatment of variousclinical pathologies.

12.4.1 The cardiac pacemaker

The cardiac pacemaker, although not considered by many to be a true motorprosthesis, is included here because the lessons learned and the technologiesdeveloped are the basis for most motor prosthetic developments. Damageor disease of the cardiac tissue can result in a number of conditions wherethe electrical impulses to contract the heart are not generated or the impulseshave failed to be conducted throughout the cardiac tissue.42 The cardiacpacemaker is an implanted device that delivers artificial electrical impulsesto the cardiac tissue to enhance or control the contraction of the heart tocorrect for these conditions.

The design of the cardiac pacemaker is similar to that described for themotor prosthesis. The stimulus delivery system is made up of lead wiresand electrodes, which can be located on the surface of the heart (epicardialelectrodes), inside the muscle of the heart (intramyocardial electrodes), orwithin the cavity of the heart pressed against the lining (endocardial orintraluminal electrodes).43,44 The lead wires travel to the control unit, whichis an implanted device located some distance from the heart in the thoracicarea. This device controls the generation of the electrical pulses delivered tothe heart based upon a sensing system that takes the place of the commandinterface of a motor prosthesis.

The lead wire design of the cardiac pacemaker is similar to the lead wiredesign in the motor prosthesis. The lead wire consists of multiple stands ofwire, helically wound, in an encapsulating sheath.43,45 The multistrandapproach to the lead wire ensures that electrical impulses can still be deliv-ered even if one of the strands breaks due to stress on the wire. Stress onthe wire, however, is reduced by the helical coil structure. This structure actslike a spring, allowing for stretching and bending, reducing breakage. Theencapsulating sheath acts to further reduce stress on the wire by allowingthe wire to move freely in the body by preventing tissue adhesions. Thesheath also provides insulation of the wire from the body, ensuring properelectrical conduction and preventing unwanted biological reactions.

The electrode in the cardiac pacemaker has to withstand repeated move-ment of the tissue and the associated stresses resulting from that movement.The electrode–lead wire interface must also be able to withstand thesestresses, especially given the fact that the stress will be concentrated at this



point. Finally, the material from which the electrode is constructed has to bebiologically compatible. The design that was developed for the electrodewas a platinum or platinum alloy disk or probe imbedded in a supportstructure constructed from the same material as the encapsulating sheath.43

The materials used ensured good biocompatibility, while the use of theplatinum stimulating electrode provided the ability to continuously excitemuscle tissue without electrode corrosion occurring. The support is contin-uous with that of the lead wire sheath, reducing stresses at the electrodewire interface and reducing the chance of breakage.

The control unit of the cardiac pacemaker is called the implanted pulsegenerator (IPG). It is constructed of a titanium or stainless steel package.43

The metallic package of the implant allows it to act, in some cases, as thereturn anode for the delivery of electrical pulses. This principle is used inmost implanted motor prostheses. Communication with the implant, forprogramming of the system or adjustment of parameters for electrical stim-ulation, is provided by a radiofrequency (RF) link. A receiver coil is locatedin the implant package, and, with an external coil placed over the implant,information can be sent to the system without the need for surgery to adjustthe system.

The cardiac pacemaker, described here, did not begin to take that formuntil the 1950s and 1960s, although the first indications of using electricalstimulation to control the heart muscle can be found as early as 1930.46 Thereason for this rapid change in the design of the system and its widespreadclinical use can be found in the introduction of the transistor in the 1940s.The transistor allowed smaller devices to be developed and eventually to beimplanted within the body. At the same time, with the advances being madein the pacemaker design, individuals began to apply the concepts of electricalstimulation to the control of the extremities.

12.4.2 Hand and foot stimulators

The first motor prosthesis for the restoration of function is the system devel-oped by Liberson and colleagues in 1961.47 The system that was developedwas used to correct for footdrop in individuals with hemiplegia. Footdrop,as is common after stroke, is the inability to lift the foot correctly during theswing phase of gait. This results in the toes not clearing the ground, causingthe individual to trip and fall. Because of this, most individuals with hemi-plegia will adopt a shuffling gait so as not to lift the foot off the ground. Thefirst motor prosthetic system developed by Liberson corrected for footdropby stimulating the peroneal nerve in the leg using surface-mounted elec-trodes. Stimulation of this nerve caused the contraction of the tibalis anteriormuscle, resulting in dorsiflexion of the foot and allowing the toes to clearthe floor. A switch was located in the sole of the shoe, which closed thecircuit when the foot was lifted off the ground and stimulated the nerve.Approximately 100 individuals were treated with this system, achievingsome level of gait improvement.



Long and Masciarelli accomplished the application of motor prosthe-ses to the upper extremity shortly after the work of Liberson.48 The systemthat was developed was used to control hand function in individualswith a cervical level injury at the fourth and fifth levels. This was accom-plished by using electrical stimulation in combination with a hand andwrist splint. A surface electrode was placed over the motor point of theextensor digitorum communis to provide for finger extension. Fingerflexion was achieved by using the splint to hold the middle and indexfingers together and force them into opposition to the thumb using aspring. The splint was also used to stabilize the wrist. A potentiometerwas mounted on the opposite arm to control the level of stimulationdelivered to the extensor muscle. As stimulation was increased, the acti-vation of the extensor muscle overcame the stiffness of the spring allow-ing the hand to open. When the stimulus level was decreased, the stiffnessof the spring dominated the muscle, and the hand closed. This devicewas tested in one individual for a period of 16 months, allowing for anincreased level of independence; however, the device did not see wide-spread clinical acceptance or use.

These first motor prosthetic systems established the fact that electricalstimulation could be effectively used to restore function to individuals withcentral nervous system injuries or diseases. However, these first systemswere quite simple in that activation was only over a single muscle, thehuman–machine interfaces relied upon simple switches or potentiometers,and surface-stimulating electrodes were cumbersome and did not providespecific activation. The challenges facing development of the motor prosthe-sis included activating multiple muscles, developing better user interfaces,advancing the technology to remove as much external hardware as possible,and broadening the user population for these devices.

12.5 Limitations of a motor prosthesisBefore continuing on to a review of current commercial and research motorprostheses, it would be beneficial at this point to review the limitations ofthe motor prosthesis. These limitations are areas of future research, and effortis ongoing to overcome these limitations to increase the clinical applicationsof these devices and to provide greater function with the existing systems.

The first limitation of the motor prosthesis, introduced briefly in thediscussion of spinal cord injury, is the problem of denervation. The motorprosthesis operates by electrically activating the nerve to the muscle. Thereason for this is that by stimulating the nerve it is possible to activateparts of the muscle not accessible by the surface. This allows for therecruitment of more motor units and the generation of greater muscle force.In addition, the stimulation of the nerve requires a smaller charge, thusless energy, than that needed to recruit the muscle fiber directly.49 Fig-ure 12.5 illustrates the differences in charge required to activate a nerve vs.a muscle fiber, as indicated by the strength–duration curve. Charge in this



case is the relationship between pulse width and stimulus amplitude,represented mathematically as:

Q = PW

× A (12.3)

where Q is the charge, measured in columbs, PW is the pulse width of thedelivered charge (measured in seconds), and A is the amplitude of thestimulus pulse, measured in amperes.

As can be seen in the graph, it requires a substantially greater chargeto activate a muscle fiber than the nerve going to the muscle. These highercharge levels are also at the point where tissue damage can occur. Tissuedamage in this case is the result of direct heating of the tissue or by thecreation of toxic chemicals due to the electrochemical reactions beingdriven at the electrode-tissue interface.50 Therefore, for these reasons motorprostheses are not used for those cases where the nerve to the muscle hasatrophied or the nerve can no longer be stimulated due to the loss of themyelin sheath.

The second limitation of the motor prosthesis is the problem of musclespasticity. Muscle spasticity is the tonic level of muscle activation that canoccur with some central nervous system injuries or diseases. In this case,action potentials are spontaneously active in the nerve, which in turn gen-erates muscle contraction, which prevents the use of the motor prosthesis

Figure 12.5 Strength–duration curve comparison of the charge required to excite anerve compared to that of a muscle fiber. The differences in charge required to excitemuscle fiber alone preclude this method as a means of restoring function with amotor prosthesis.

Pulse Width ( msec )

Am

plitu

de (

mA

)

0.01 1.0 10.0 100.00.1

MuscleFiberNerve

20

40

80

100



for the restoration of function because the current designs rely upon gener-ating action potentials in the nerve to control contraction. However, giventhe ability of the motor prosthesis to generate an action potential, it can beseen how this can be applied to the problem of spasticity. The action potentialgenerated on a nerve by the motor prosthesis travels in both directions fromthe stimulation site, both down to the muscle and back toward the spinalcord. However, with the nerve cuff electrode, it is possible to generate anaction potential in only one direction.51 If an action potential is travelingfrom the central nervous system to the muscle at the same time the artificiallygenerated action potential is traveling up the nerve, both potentials will meetand cancel each other out. This principle, referred to as collision blocking(Figure 12.6), can be used to control spasticity and could be applied to someof the injuries and diseases discussed earlier.

The final limitation of the motor prosthesis is the limited number offeedback signals available. In Figure 12.3, the system design of the motorprosthesis indicated performance and environmental feedback being sent tothe user. However, in most motor prosthesis applications, this feedbackinformation is limited to visual and auditory feedback, with little or noinformation from sensory receptors (e.g., touch, temperature, pain) or muscleproprioceptors (e.g., joint position, force).52,53 The user of the motor prosthesismust be continually monitoring the output of the system visually, and relyupon whatever machine information is being sent back through auditory

Figure 12.6 The principle of collision blocking. (A) The natural progression of anaction potential from the neuron down the axon. (B) With the application of anelectrical pulse, action potentials can be generated in both directions. (c) With a nervecuff electrode, pulses can be generated so that they travel up the neuron toward thedirection of the cell body so that they can block any impulse traveling down the nerve.

(A)

(B)

(C)



and visual cues. This places a high cognitive demand upon the user of sucha system, which makes the motor prosthesis, at times, difficult to use andimplement. Several methods have been explored to provide sensory feed-back though electrical stimulation of areas where sensory function isintact54,55 or by recording from the nerves going to cutaneous receptors andthen passing this information onto the motor prosthesis.56,57 However, theissue of feedback to the user is one that has yet to be addressed in a satis-factory manner.

12.6 Commercially available motor prosthesesThe results achieved with the first motor prostheses, and the clinical successof the cardiac pacemaker, have led to the development and marketing ofseveral motor prosthetic systems for clinical use. Figure 12.7 shows the clin-ical deployment of two of the commercially available motor prosthetic sys-tems in recent years. Although the field of motor prosthetics is relativelynew, it can be seen that after their introduction there has been an increasingapplication of motor prosthetics in the rehabilitation field. If this curve isprojected out at its current rate, there will be more than 100,000 motorprostheses in use by the year 2010. Although only two motor prostheticsystems are shown in Figure 12.7, several are commercially available for use.The goal of this section will be to review these systems. In order to facilitatethis discussion, the commercial systems will be divided into three groups

Figure 12.7 Plot of number of commercially available motor prostheses implantedsince 1976. Only the information about two of the systems, one for the upper extrem-ity and one for bladder and bowel restoration, are shown.

1

10

100

1000

10000

1976 1981 1986 1991 1996

Num

ber

of Im

plan

t Rec

ipie

nts

Vocare System

Freehand Implant



based upon their function: those for upper extremity movement, those forlower extremity movement, and those used for organ systems.

12.6.1 Upper extremity motor prostheses

Four motor prostheses are commercially available for the restoration of func-tion in the upper extremity. Most of these systems are directed towards therestoration of hand function after a spinal cord injury; however, these sys-tems have also found applications for motor relearning after stroke, and oneis specifically marketed for this purpose.

The first system to be reviewed is the Bionic Glove, also known as theTetron Glove, originally marketed by NeuroMotion, Inc.58,59 This device wasdesigned for individuals with a cervical level injury at the sixth and seventhlevel to augment the natural tendonesis grasp to provide active hand func-tion. Individuals with this level of spinal cord injury will have active controlover shoulder and elbow movement and active wrist flexion. However, thereis no remaining control over the muscles for the hand to provide for grasp.Therefore, they rely upon the use of the tendonesis grasp to acquire objects.Tendonesis grasp is a naturally occurring phenomenon due to the length ofthe finger flexor tendons. The primary finger flexors are in the forearm, andthe tendon for these muscles crosses both the wrist and the finger joints. Asthe wrist is extended (pulled up against gravity), tension on the finger flexortendons increases, causing the tendon to pull the fingers into flexion. Thisgrasp, while allowing the hand to close, is often insufficient in force to allowan individual to grasp an object.

The Bionic Glove augments this grasp by stimulation of the finger andthumb flexors and the finger extensors using surface-mounted electrodes.The system consists of three electrodes that are placed over the motorpoints of the muscles to be stimulated, with the return anode locatedproximal to the wrist. Metal studs are located on the back of each electrode,which can connect to a stainless steel mesh located inside a neoprene glove.When the glove is worn, the studs interface with the mesh, completing thestimulus delivery system. The control unit for the system is located on theforearm portion of the glove and provides for the electrical activation ofthe muscles. A position transducer attached to the wrist provides controlover the unit. As the individual flexes the wrist, the finger extensor musclesare stimulated to provide hand opening. When the wrist is extended,stimulation to the extensors is decreased and the stimulation of the fingerflexors is increased.

The Bionic Glove achieved limited clinical success, being implementedin 37 individuals by 1996.59 The device was found to improve function duringactivity of daily living (ADL) assessments, providing the ability to grasp andmanipulate larger and heavier objects could be accomplished without thesystem. The exercise regimen used to strengthen the atrophied muscles alsodemonstrated therapeutic benefits in that individuals, after several monthsof using the system, no longer required it for use in ADL tasks.58



The second commercially available system also uses a surface electrodedelivery system for the restoration of hand grasp. This device is the Hand-master system, marketed by Neuromuscular Stimulation Systems, Ltd.30,60–62

This device is designed for individuals with a cervical-level spinal cordinjury sustained at the fifth level and for individuals with hemiplegia. Thedevice consists of three to five electrodes used to stimulate the finger flexorsand extensors and the thumb flexor muscles. However, unlike the BionicGlove, these electrodes are permanently mounted in an orthotic brace. Thismakes the system easier to use than the Bionic Glove because individualelectrode placement is not required each time the system is used. The orthoticbrace also prevents movement at the wrist, but this stability is essential inindividuals with a C5-level injury who have completely paralyzed wristflexors and extensors.

The control unit of the system is external and independent from thebrace, with a belt clip to allow the unit to be worn about the waist. Controlover the application of the electrical stimulation is accomplished by a seriesof switches and a potentiometer. The switches are used to electrically stim-ulate either the finger flexors or extensors to control the degree of handopening and closing, with the potentiometer used to control the degree ofthumb flexion. Two additional push-button switches allow the user toincrease the level of grasp force.

The location of the switches and the potentiometer are on the controlunit and require the use of the opposite hand for control. While this arrange-ment may not be ideal for individuals with a spinal cord injury, it is appli-cable for individuals with hemiplegia. The Handmaster system, in fact, hasbeen proven to be a valuable clinical tool for these individuals. Results insubjects tested with this system have indicated a reduction in spasticity asmeasured by improved ranges of motion about the hand and wrist joints,and better hand function.60 This device is currently available in Europe, withapproval by the Food and Drug Administration (FDA) to begin marketingand sales in the United States.

A third commercially available system is the AutoMove AM800, marketedby Stroke Recovery Systems, Inc.63 This device, like the others, uses surfaceelectrode technology to activate the muscles. However, unlike the othersystems, this device is marketed solely as a motor recovery device for indi-viduals who have sustained hemiplegia. The device uses three to five elec-trodes to activate the muscles of the wrist and hand. Control over the appli-cation of the stimulation is accomplished by the recording of muscle activityfrom the same electrode set. When the attempt to active a muscle is recorded(i.e., a twitch in the EMG signal), the command is sent to the control unit toactive the muscle electrically. In this manner, the motor command from theindividual is augmented by electrical stimulation. This makes it possible forthe brain to relearn how to activate the muscle, reducing the effects ofhemiplegia. The electrical activation of the muscles also reduces spasticityin these individuals. This device is currently available in the United Statesand is FDA approved.



A final motor prosthetic system that is commercially available is theFreehand System, marketed by NeuroControl Corp.64–68 This device, unlikethe other devices discussed, is an implanted system aimed at restoring handfunction to individuals with a cervical level spinal cord injury at the fifth orsixth level. The device consists of an implanted stimulator, located in thechest, with lead wires for eight stimulating electrodes. These lead wires aretunneled under the skin and are connected to implanted electrodes locatedin or on the motor points of the muscles of the forearm and hand. Thestimulation parameters, and the power to the implanted device, are providedthrough a RF coil placed on the skin over the implant. The external controlunit, which contains the information on the muscle activation patterns andthe batteries, communicates with the implant via the coil.

The Freehand system provides the individual with two grasp patterns,a palmar grasp and a lateral grasp. In the palmar grasp, the thumb is broughtinto opposition to the fingers and the fingers are flexed against the thumb.This grasp is useful in acquiring and holding objects such as drinking glassesand books. In the lateral grasp, the fingers are flexed and the thumb is flexedagainst the side of the index finger. This grasp is useful is acquiring andholding objects such as pencils, forks, and keys.

The user of the Freehand system has complete control over the degree ofhand opening and closing, as well as the selection of grasp, through the useof a transducer mounted on the shoulder opposite to the arm implementedwith the Freehand system. The shoulder transducer consists of two parts, apush button switch and a joystick. The switch is used to operate the binaryfunctions of the system, such as grasp selection and turning the system onand off. The joystick, which spans the glenohumeral joint of the shoulder,provides the input signal to control hand opening and closing. The shoulderrange of motion is mapped directly to the degree of hand opening and closing.By moving the shoulder through the range of motion, the user can positionthe hand at any point between full hand opening and full hand closure.

The Freehand system has achieved a modest level of success, beingimplemented in over 200 individuals worldwide. The device has been dem-onstrated to improve grasp force in individuals with sustained tetraplegia,to improve performance on ADL assessments, and to improve hand postureand joint range of motion.67,68

12.6.2 Lower extremity motor prostheses

Currently, there are five commercially available motor prosthetic systems forthe restoration of function in the lower extremity. However, given the com-plications in the biomechanics of the lower extremity, the focus has primarilybeen on providing systems that correct for footdrop in individuals withsustained hemiplegia or for providing standing in individuals with sustainedparaplegia. Walking has been achieved to a limited extent; however, thesesystems are dependent upon the use of a walker or crutches in conjunctionwith the system to provide upright stability.



Five systems are currently available to correct for footdrop: WalkAid,originally marketed by NeuroMotion, Inc.;69,70 Odstock Dropped Foot Stimula-tor (ODFS), made available for use by the Department of Medical Physicsand Biomedical Engineering, Salisbury District Hospital, United King-dom;71,72 MicroFES, made available for use by the Institute for Rehabilitationin Ljubljana, Slovenia;73,74 Footlifter, marketed by Elmetec A/S;75 and PNS2000, marketed by BarMed Pty., Ltd. All of these systems provide for theelectrical activation of the muscle with surface-mounted electrodes. Thesesystems consist of one or two electrodes, placed over the peroneal nerve tocause foot dorsiflexion and knee flexion (withdrawal reflex). An externalcontrol unit is mounted on the calf or above the knee using Velcro straps.Control over the system is provided by either a foot-mounted switch torecord when the foot is lifted off of the ground or a push-button switch tocontrol the application of stimulation. These systems can also be placed intoan exercise mode to improve muscle strength and reduce muscle spasticity.These devices have experienced a modest level of clinical acceptance, with7500 to 10,000 devices currently in use.75

The Parastep system,26,76,77 marketed by Sigmedics, is the only commer-cially available system for the restoration of standing and walking in indi-viduals with sustained paraplegia. This device has been implemented inover 600 individuals to date.78 The device consists of an external control unit,surface electrodes, and push-button switches that are mounted on a walker.This system is to be used in conjunction with the walker to provide uprightstability and to move the individual forward. Six electrodes are used in thissystem, four to provide upright standing and two to elicit rudimentarystepping. Standing in the system is provided by the electrical activation ofthe quadriceps muscles to fix the knee in extension and activation of theparaspinals and the gluteal muscles to provide lower back stability and hipextension. Stepping is achieved by the stimulation of the peroneal nerve, asin the footdrop systems, to activate the withdrawal reflex. Control overstepping is achieved by the push-button switches, one for each leg. Whenthe button is pressed, the leg is stimulated to achieve knee flexion and ankledorsiflexion, while the quadriceps of the other leg are stimulated to acceptthe weight of the individual. The control unit for this system is worn aroundthe waist of the individual and contains the power source and the electricalactivation patterns for each individual.

12.6.3 Organ system prostheses

The commercial applications of motor prostheses for the restoration of bio-logical function are focused on the restoration of bladder and bowel functionand in the restoration of respiratory function in those individuals with ahigh cervical level spinal cord injury. Four systems have been identified foruse in bladder control and/or bladder continence: Brindley–Finetech system,marketed by NeuroControl Corp.; LEVATOR Turbo CS200 Continence Stimu-lator, marketed by Ferraris Medical, Ltd.; Interstim, marketed by Medtronic,



Inc.; and a stimulator marketed by Dobelle Institute Avery Laboratories. Allof these stimulators are available in the United States and generally operatealong the same principles, stimulation of the sacral nerves to control bladderand urethral contraction.

The Brindley–Finetech Bladder Controller, also known as the VOCAREsystem in the United States,79,80 consists of two electrode pairs and a stimu-lator implanted within the body and an external control unit worn aroundthe waist. The electrodes are implanted near the second through fourth sacralnerves bilaterally in the sacral canal after they have been exposed by alaminectomy. The stimulating electrodes provide for the contraction of thebladder wall to induce urine flow. The location of the electrodes on thesecond through fourth sacral nerves also provides for contraction of thelower bowel and relaxation of the anal sphincter in approximately 50% ofthe recipients of the system, promoting regularity and greatly reducing thetime of the bowel program.3 Control over the system is maintained by theuser, with a switch selector on the external control unit to allow the individ-ual to control micturition and defecation.

The system has achieved a great deal of clinical success, being imple-mented in over a thousand individuals worldwide.3 Implementation of thissystem is also frequently accompanied by a dorsal rhizotomy of the secondthrough fifth sacral nerves. This improves continence and prevents invol-untary urethral sphincter contraction with the electrical contraction of thebladder wall. The problem with this, however, is a loss of reflex erection,ejaculation, perineal sensation, and an alteration in reflexive defecation.These drawbacks have somewhat limited the acceptance of this system,primarily by males.

Four systems are commercially available for use in the restoration ofrespiration or to provide for coughing in individuals with high cervical levelspinal cord injuiries: Quik-Coff system, marketed by B&B Medical Technol-ogies outside of the United States; Diaphragm Pacing system, marketed byDobelle Institute Avery Laboratories; Atrostim system, marketed by AtrotechOy, Finland; and T154, marketed by MedImplant, Austria. These devices,except for the Quik-Coff system, are implanted devices that stimulate thephrenic nerve to cause contraction of the diaphragm to restore respirationwithout ventilatory support. These implanted devices, except for the T154,are available in the United States for use in high tetraplegia.

The Quik-Coff system is a surface-electrode system developed to assistindividuals with cervical level spinal cord injuries with coughing.81 Surfaceelectrodes are placed on the abdominal wall, in conjunction with an abdom-inal binder, to activate the external and internal oblique muscles. An indi-vidual who wishes to initiate or needs assistance with a cough uses a push-button switch to activate the muscles. Coughing can be accomplished ona single cough attempt or multiple attempts (up to four) separated by 1-,2-, 3-, or 4-second intervals. The control unit of the system is a small,battery-powered device the size of a pager that can be worn around thewaist with a belt clip. Results with this system are encouraging, indicating



that expiratory pressure can be increased up to two and a half times ofwhat can be achieved with no assistance.81 This can greatly reduce pulmo-nary complications associated with spinal cord injury.

12.7 Current research motor prosthetic systemsOngoing studies in the area of motor prostheses are focused at expandingthe current applications of the systems. This is achieved by providing formore channels of activation to recruit more muscles, improving the electricalstimulation technology, and expanding the clinical indications for motorprostheses. This section will attempt to provide an overview of currentresearch in this area, again dividing the systems into those for use in theupper extremity, the lower extremity, and for the restoration of organ systemfunction.

12.7.1 Upper extremity motor prostheses

Several investigators are working in the area of upper extremity motorprostheses with the goal of either providing for a greater restoration offunction or improving motor prostheses technology. One series of systemsaimed at providing both greater function with the motor prostheses andadvancing motor prostheses technology are the implanted stimulator tele-meters (ISTs) being developed at Case Western Reserve University. Thesenewer systems are an expansion of the Freehand system mentioned earlierand are the first implanted, sensor-driven motor prostheses ever to be devel-oped. The IST systems provide for a greater restoration of function by allow-ing more muscles to be incorporated into the electrically stimulated handgrasp, up to a maximum of 16 muscles. Examples of improved functioninclude the fact that individuals using these systems are provided withelectrical stimulation of the triceps muscle to allow for active elbow flexionand extension, which can greatly increase their usable workspace.82 In addi-tion, these individuals are provided with electrical activation of the fingerintrinsic muscles,83 which has improved grasp strength and helped maintaina more natural hand posture.

The technology of the IST systems is also an improvement upon motorprosthesis technology in that the RF link with the implanted system providesfor two-way communication. This has allowed for the implantation of thecommand interface for the users of this device. The implanted joint angletransducer (IJAT) available with this system is a Hall effect sensor that isimplanted into the bones of the wrist.24,55 As the individual moves their wristin flexion and extension, the IJAT system provides an output of the jointangle that is used to control the degree of hand opening and closing in thesame hand. The IST system also allows for electrodes to be placed on musclesunder voluntary control and uses the EMG signal to provide control overthe hand grasp55 in those cases where the individual does not have activewrist extension.



Another system designed at providing greater function with motorprostheses and advancing motor prosthesis technology is the system devel-oped at Tohoku University in Japan.27–29,84–88 This system is designed as amultipurpose system that can be used to restore complete upper extremityfunction regardless of the level of injury, to provide for standing andwalking, and for therapeutic use. The system provides for up to 64 channelsof stimulation delivered through the use of percutaneous electrodes. Con-trol over this system is provided either with a respiration (sip-puff) con-troller or through voice recognition software. Stimulation patterns to bedelivered to the muscles are based upon EMG activation patterns recordedduring the accomplishment of different tasks in able-bodied individu-als.84,86 This device has been demonstrated to be useful in the restorationof function in individuals with a cervical level injury at the fourth levelwhere shoulder and elbow control are also required and for walking inindividuals with thoracic level injuries.

The disadvantages of the research systems described above are that theyall involve placement of materials inside the body that may be consideredunacceptable by some of the end users of these devices. Therefore, someinvestigators are focusing on the development of surface stimulation systemsfor use in the upper extremity. The ETHZ-ParaCare system, now being mar-keted in Europe as the Compex Motion system, is a new system that providesfor four channels of stimulation using surface-mounted electrodes.78,89 Thecontrol unit device is a small, pager-sized system that can be clipped ontoa belt or placed inside a pocket, with overall control of the system by theuser being provided by electromyographic signals under voluntary control.The control unit can be programmed using a graphical interface, and theinterface and the memory cards inside the control unit are interchangeable.This allows the system to be individually tailored to each individual andapplied quickly and easily. The four channels of stimulation in the systemcan be expanded in multiples of four, although how this is accomplished orthe maximum number of available channels has not been reported. Cur-rently, ten individuals have used this system for walking, hand grasp, andtreatment of shoulder subluxation.90

A final series of investigations into the uses of motor prostheses for theupper extremity are those that have examined the effects of electrical stim-ulation on motor relearning.5,60,91–94 In addition to shoulder subluxation, theeffects of hemiplegia on motor control can include the inability to activatenatural movement patterns in the upper extremity to interact with the envi-ronment, as well as paralysis of certain muscles. Investigations at severaluniversities and institutions.91–94 are directed toward systems that can recordtrace activity of muscles in the forearm and hand and then apply electricalstimulation to the same muscle to achieve full muscle recruitment. This isthe same concept as discussed in the section on the AutoMove AM800. Byamplification of muscle activity, the goal is to retrain motor cortical areasusing positive feedback mechanisms. This should provide for a more efficientrecruitment of the muscle to achieve grasping and movement.



Other investigators are also addressing paralysis and spasticity in thehemiparetic hand.5,60,89,91 A common occurrence with hemiplegia resultingfrom stroke is the inability to active the finger flexors to achieve hand closure,in conjunction with a tonic level of activity in the extensor muscle groupsof the forearm in hand. These systems are designed to address these twoneeds separately to provide for efficient hand grasp. The principle of collisionblocking, as discussed earlier, is used in some systems to prevent the toniclevel of contraction for the finger extensors by preventing the conduction ofthe action potential along the motor neuron to the extensor muscle. Whenfinger extension is required, these systems turn off the blocking pulses,allowing finger extension to occur. When finger flexion is required, thesesystems then block the action potentials to the finger extensors while at thesame time stimulating the finger flexors to close the hand. Other systemsaddress this problem by increasing the stimulation to the finger flexorswithout blocking the extensors, with the idea that the flexors are strongerthan the extensors and will overpower them to allow for hand closing. Thedifficulty with many of these systems, however, is that stimulation isachieved through the use of surface or percutaneous electrodes, both ofwhich have been unacceptable to the end user of the device because of theunpleasant sensation elicited by the electrical stimulation.

12.7.2 Lower extremity motor prostheses

Current research trends for the lower extremity motor prosthesis, such asin the upper extremity, are directed toward increasing the functional capa-bilities of the motor prosthesis and advancing the technology. To this end,several different investigations have been identified that are aimed atimproving walking and standing using motor prostheses. The researchbeing conducted at the Salisbury District Hospital95 is directed towardproviding standing in individuals with sustained paraplegia using sacralroot stimulation. The system that has been developed, the lumbosacralanterior root stimulator implant (LARSI), uses 12 intradural electrodesplaced on the second lumbar through the second sacral anterior roots inthe cauda equina. Postsurgical stimulation of each of the roots individuallyis performed to identify joint movement generating capabilities and nervestimulation combinations to achieve upright standing. Currently, two indi-viduals have received the system and, even though standing is possible,they have encountered difficulties in attaining a good posture due to exces-sive hip flexion occurring with the sacral root stimulation. Current researchby this group is underway to reduce hip flexion to achieve upright, hands-free standing.

The Praxis-24 system being developed by Neopraxis Pty., Ltd.96,97 is animplanted system that provides for up to 22 channels of electrical stimula-tion. This system is designed to provide for upright standing and walkingfor individuals with sustained paraplegia, in addition to providing for blad-der control. Electrodes placed adjacent to the motor nerves in the quadriceps,



hamstrings, ankle, and gluteal muscles provide standing and walking func-tions. Additional electrodes are placed near the motor nerves for the psoasmuscle to provide for hip flexion. Control over the system is accomplishedusing gyroscope and accelerometers to record limb position, with overallcontrol being maintained by the user through a touch-sensitive LCD screenon the control unit.

Bladder function in this system is achieved by the placement of threeelectrodes near the first through third sacral spinal nerves, exposed by alaminectomy. Another electrode is placed over the conus medullaris at thetwelfth thoracic/first lumbar level to remove the need for the dorsal rhizo-tomy. This system has currently been implanted in two individuals with amodest degree of success. These individuals have been able to achieve stand-ing while being able to manipulate objects with one hand. In addition, thesystem has been able to provide bladder emptying in one subject withpressures of 50 to 70 centimeters of water.97

Another system that is directed toward walking and standing in indi-viduals with sustained paraplegia is the 16-channel neural/epimysial stim-ulator that is being developed under the European “Stand-Up-And-Walk”(SUAW) project.98 This work is early in the research phase and limited infor-mation is available; however, the results have been encouraging and warrantmention here. The device is an implanted system using a combination ofmuscle- and nerve-stimulating electrodes, with overall control over walkingbeing maintained by the individual with push-button switches. This devicehas been implemented in two individuals, with one individual being ableto achieve standing and short distance walking, while the other individualis still in the training period for the device.

A final system under development for standing and transfers in indi-viduals with sustained paraplegia is the implanted system under develop-ment at Case Western Reserve University.99,100 The implantable receiver usedin this system is identical to the early upper extremity system that led to thedevelopment of the Freehand system. The system provides for eight channelsof stimulation, delivered by the use of epimysial and intramuscular elec-trodes. The electrodes are placed bilaterally in the vastus lateralis, gluteusmaximus, and semimembranosus muscles to provide for knee and hip exten-sion. The other two electrodes are placed near the twelfth thoracic and firstlumbar nerve roots to provide for stimulation of the erector spinae to providetrunk extension. Control over the system is maintained by the user with apush-button switch, independent of the control unit for the device, which isa portable unit worn around the waist.

This system has been implemented in 13 subjects to date,100 all of whomhave been able to achieve upright standing with the use of a walker. In mostcases, the legs have supported approximately 80 to 95% of the body weight,allowing the users to perform limited, one-handed manual tasks.101 Ongoingresearch is being directed toward achieving longer standing times in thissystem before muscle fatigue requires the individual to turn the system off.Work is also directed toward trying to achieve the distribution of all of the



body weight to the legs to allow for hands-free standing and limited mobilitywith the use of a walker.

The clinical research on the standing and walking motor prostheses andthe advances being made there are mirrored by the ongoing research aimedat correction of gait in individuals with hemiplegia. At least three series ofinvestigations are ongoing to correct for footdrop in hemiplegia. Two of theseresearch systems, like the commercially available systems, rely upon stimu-lation of the peroneal nerve to elicit ankle dorsiflexion. However, the stim-ulation and control methods are different. The system developed at theUniversity of Aalborg102 uses nerve cuff electrodes to record activity fromthe sural nerve which innervates the heel region. By monitoring activity inthis nerve, it has been possible to record heel strike and the toe-off phase ofgait. This signal has then been used in place of the external switches usedin the other systems to provide stimulation to the peroneal nerve whenrequired. The systems being developed in Ljubljana103 have focused moreon the implantation of the stimulating electrodes, with placement of theelectrode on the peroneal nerve to avoid issues with electrode placementand unwanted recruitment of sensory nerves in the skin. This system hasalso been expanded to stimulate other muscles in the leg to provide forstanding and walking for individuals with sustained paraplegia.104,105

The third clinical system under development is the footdrop systemunder investigation at the Cleveland FES Center.106,107 This system differsfrom many of the others discussed in that stimulation of the muscles tocorrect for foot drop is not achieved by stimulation of the peroneal nerve.Instead, percutaneous electrodes are used to individually excite muscles ofthe lower leg, as well as those around the knee and hip, to improve gait. Inaddition, studies with this system have examined the effect of electricalstimulation on motor relearning in the lower extremity. The results of thissystem are better than what has been achieved with the other systems inthat stimulating the individual muscles instead of the peroneal nerve to elicitthe withdrawal reflex eliminates concerns about accommodation to the stim-ulation over time.108

12.7.3 Organ system prostheses

Although the commercial applications of motor prostheses for the restora-tion of organ system functions have been limited at this time, there arenumerous investigations ongoing into using motor prostheses for the res-toration of these functions that should result in a variety of commercialsystems in the future. One series of investigations is directed toward usingmotor prostheses for the treatment of upper airway disorders. Research atthe National Institutes of Health109–111 is directed towards the developmentof motor prostheses that can be used to stimulate the genioglossus nerveto open the hypopharynx to correct for sleep apnea. Sleep apnea is a con-dition in which the tongue muscles relax during sleep, resulting in restric-tion and blockage of the hypopharynx. Studies are also being conducted on



stimulation of the thyroarytenoid muscles to prevent aspiration in dysph-agia and on stimulation of the thyroarytenoid muscle to correct for spas-modic dystonia, a voice disorder that prevents effective communication.

Another area of investigation for the use of motor prostheses for therestoration of organ system function is in the area of cardiac assist devices.The cardiac assist device differs from the cardiac pacemaker in that skeletalmuscle is used to repair or reinforce cardiac muscle. The skeletal muscle isconditioned to convert the fibers to a fatigue-resistant variety and is electri-cally stimulated to contract when the heart contracts to increase blood pres-sure and blood flow. The most common muscle used is the latissimus dorsi,which is removed from its origin on the back and hip, inserted into the chest,and wrapped around the heart. Investigations at the Milwaukee HeartInstitute112 have been directed toward the development of a motor prosthesisthat can sense the contraction of the heart and then provide the stimulationof the latissimus dorsi to assist in the contraction. This device, called the LDPace II, has undergone successful animal testing and has been implanted inone human subject to date.112

The use of motor prostheses for the restoration of respiration after spinalcord injury is another area of investigation. Numerous studies have inves-tigated the possibility of electrically stimulating the phrenic nerve to elicitcontraction of the diaphragm and restore breathing.113–116 The difficulty inusing these systems, however, are that individuals are reluctant to undergosurgery for a system that may damage the nerve. Research being conductedat Case Western Reserve University has addressed this concern by develop-ing two different motor prosthetic systems.117,118 The first system electricallystimulates the intercostal muscles using intramuscular electrodes to providefor respiration in the cases where the phrenic nerve is lost or damaged orwhere surgery is undesired by the individual. The second system alsoemploys the use of intramuscular electrodes, but these are placed on themotor point of the diaphragm. By using these approaches, these researchsystems have been able to successfully address the problems of phrenicstimulators developed in the past. The diaphragm stimulator has alreadybeen successfully implanted in one human subject.118

A final area of investigation for the use of motor prostheses for restorationof biological functions is in the area of the restoration of bladder, bowel, andsexual functions. The objective of these studies has been the development ofsystems that restore all of these functions and accomplish this without theneed for a dorsal rhizotomy. This involves primarily stimulation of the puden-dal afferent pathways to prevent incontinence and to increase bladder volume.Studies at the University College London119,120 and Case Western ReserveUniversity121 show promise in the use of this method to avoid the dorsalrhizotomy, with four individuals being implanted with the sacral posteriorand anterior root stimulator implant (SPARSI) at the University College Lon-don. Recipients of the SPARSI system have been shown to have a significantreduction in reflex incontinence with stimulation of the second through fourthsacral roots120 with accompanying increases in bladder volume.120



12.8 Future avenues of investigationThe previous sections of this chapter have attempted to present the clinicalapplications of motor prosthetics and the systems that have been or are beingdeveloped to address these needs. As can be seen from these previous sec-tions, a great deal of work has been accomplished in the area of motorprosthetics; however, it is far from being complete. The clinical need toincrease the user base of these systems not only warrants investigations intohow to apply these systems to address central nervous system pathologies,but also investigations in how to improve the technology and functionalityof these systems. This section will attempt to introduce where work is juststarting or needs to be accomplished in order to address the clinical needsfor these systems.

12.8.1 New clinical applications

The previous sections on the commercial and research motor prostheses mayleave the reader with the false impression that motor prostheses are onlyeffective for use in individuals with a spinal cord injury, with some minorapplications to individuals recovering from stroke. The reason for this one-sided application of the motor prosthesis, however, lies in one of in thefundamental limitations of these systems. Current technologies for the sys-tem rely upon an intact motor neuron to the muscle in order to generate amuscle contraction. If the nerve is damaged or lost, then the usefulness ofthe current motor prosthesis technology for the treatment of central nervoussystem deficits is lost.

The limitation that an intact motor nerve must be present in order togenerate a muscle contraction has limited the application of motor prosthe-sis technology in the areas of neuropathy, amyotrophic lateral sclerosis, andother diseases where the myelin sheath of the nerve has degenerated andthus the nerve is no longer capable of conducting an action potential orwhere the nerve fibers themselves are attacked by the pathogen and thuslost. However, in these conditions, there is a definite need for a device thatcan restore functional movement or biological function. Solutions to thisproblem may lie in the areas of nerve and cellular regeneration. Even if fullnerve regeneration cannot be accomplished so that the connection betweenthe brain and the muscle is restored, the ability to regenerate nerves orreinnervate muscle can be extremely beneficial. If a nerve can be restoredto a muscle, even if no longer connected to higher nervous system functions,the motor prosthesis can be used to excite this nerve to cause the muscleto contract.

Motor prostheses can also be used, although not in a traditional sense,to aid in the area of neural regeneration in the spinal cord. It has beensuggested122–124 that the motor prosthesis, or more importantly the electricalcurrents generated by the motor prosthesis, can be used to guide axonalgrowth for regenerated nerves so that they make functional connections



between the central and peripheral nervous systems. Motor prostheses couldalso perform an important role in maintaining muscle mass and strengthwhile nerve regeneration was occurring so that, once the connection wascompleted, the individual would not have lost the ability to control theextremities in a functional manner, requiring months of therapy to rebuildmuscle mass and tone.

Another possible role for motor prostheses, which has been suggestedseveral times in this chapter, is in a role of reducing muscle spasticity thatcan occur in cases of sustained hemiplegia or cerebral palsy. The methodof collision blocking, introduced earlier in this chapter, has been demon-strated to be theoretically possible to prevent action potential conduction.The goal with a motor prosthetic system, therefore, would be to make thistheory a reality in a working clinical system. Such a system not only wouldhave to prevent the conduction of unwanted action potentials but wouldalso have to allow muscle contraction to occur when the user desired togenerate a movement.

12.8.2 Technology development

The hypothetical clinical applications of the motor prosthesis, and evencurrent ongoing research, will require further advances in the technology ofthe motor prosthesis in order to accomplish these goals. One area that willrequire attention will be on the development of implanted components.Although research is being directed at this time on non-implanted, non-invasive systems, future directions in motor prosthetic development willrequire the use of implanted technologies. This will be necessary in order toaddress the cosmesis and ease of use requirements for the system. A seriesof external devices that may draw unwanted attention to the user or thatwill require a long period to don and doff will most likely lead to rejectionof the device. However, the increasing amount of technology that will berequired to be implanted into the body will require a great deal of surgicaltime and increase the risk to the recipient of the system. Technologies willbe required that allow for placement of the devices into the body quicklyand easily.

The BION system, developed at the AE Mann Institute for BiomedicalEngineering, is aimed at advancing the technology of the motor prosthesisby addressing these needs.125–127 The key component of the system is theBION, which is a small, injectible, single-channel stimulator. This stimulatoris encased in a ceramic package, with the power and the simulation param-eters being provided to the BION by a RF link. The objective of this systemis to remove much of the implanted technology and lead wires associatedwith the other systems and instead have all communication be conductedby the RF link. This system has been reported to be capable of providing upto 255 independent channels of stimulation and has already been appliedfor uses in shoulder subluxation.127 Shoulder subluxation is a conditioncommon in hemiplegia where the paralysis of the rotator cuff muscles results



in partial dislocation of the humeral head, resulting in pain and an inabilityto use the shoulder effectively. This system has also been applied to strength-ening of the quadriceps muscle in individuals with osteoarthritis, althoughtests are still in the early clinical stage127 and with stimulating the pudendalnerve to treat overactive bladder.127

The demands on the development of implanted technologies for motorprosthesis user will also be complicated by the fact that these systems willbe required to restore greater and greater function. This places demands onthe amount of hardware that is implanted and raises issues on how thesedifferent components of the system will communicate. For example, if thetechnology used by the Freehand or Praxis system is developed further toallow for up to 32 or 64 channels of stimulation, this will result in numerouslead wires that will need to be tunneled under the skin, with all of themoriginating from a central processor, an arrangement that will be highlyundesirable to the physician who has to implant such a system. Likewise,external systems will have multiple lead wires running from various areason the individual to a central processor, which will most like result in a lackof cosmetic appeal. These issues will most likely be addressed through thedevelopment of modular, implanted systems that will be able to communi-cate with a limited need for wires or antennas.

The increasing complexity of these newer systems, with more functionsand more components, will place greater demands upon the power supplyto these systems. Although battery technology is developing due to advancesin telecommunications and computer science, these batteries are notdesigned with biological issues in mind. Most are not biocompatable, nordo they meet the voltage and current requirements to allow for musclestimulation for periods of months to years. Battery technology will need tobe directed toward the specific goals of biocompatibility and the ability toreplace the battery easily in the implanted system or to allow for quick andefficient recharging. One possibility for the development of a power sourcefor use in the human body are the developments in the area of biofuel cells.128

These cells will make use of the body’s own chemistry and structure, withthe introduction of some components, to provide the power to drive thesesystems. Research in this area is evolving and realization of these systemsin vivo has yet to be accomplished.

The increasing complexity of newer motor prosthetic systems will alsorequire increasing complexity in the command interface. Informationrequirements for the system will increase, with subsequent decreases in thenumber of input channels available. Generating these high information con-tent signals will also have to be accomplished while at the same time ensur-ing that the interface is easy and understandable to use. The most likelyavenue of investigation for command interfaces that will be able to meet theneeds of the complex system will be to derive command signals from thecortex. Numerous investigators are examining the feasibility of extractingcontrol information from the motor and pre-motor areas of the brain.129–131

These studies have resulted in the control over a robotic arm in primate



studies. The animal in this case moves their arm and the robot mirrors thismovement. These studies are instrumental in demonstrating that by record-ing from just a few neurons in the motor and pre-motor areas, it is possibleto reconstruct arm movement. Areas of future investigation with these stud-ies will need to address the ability to think of movement and have a systemrespond instead of having the system mirror movement. Issues of biocom-patibility and human implantation need to be explored, as do ways of allow-ing cortical signals to control not only functional movement but system state(e.g., turning the system on/off, selection of exercise modes), as well.

12.9 Summary and conclusionThe objective of this chapter was to provide the reader with a sufficientbackground in the area of motor prosthetics in order to allow them to beginto conduct their own research in this area. To this end, a review of past motorprostheses was presented, as well as an overview of the technology that hasbeen employed to date. As can be seen from this review, motor prostheseshave achieved a great deal of clinical success; however, they are stillunderutilized by rehabilitation practitioners. In order for motor prostheticsto continue to develop as a field, it is necessary for research to continue intoexpanding the clinical applications of these devices, to continue to developthe technology so that it becomes an integral part of the individual for whichthe system is designed, and to educate individuals as to what has beendeveloped and what is currently in the research stages so that these devicescan see greater clinical use.

References1. Heimer, L., The Human Brain and Spinal Cord, 2nd ed., Springer-Verlag, New

York, 1995, chap. 24.2. Peckham, P.H., Mortimer, J.T., and Marsolais, E.B., Alteration in the force and

fatigability of skeletal muscle in quadriplegic humans following exercise in-duced by chronic electrical stimulation, Clin. Orthop., 114, 326, 1976.

3. Creasey, G., Restoration of bladder, bowel, and sexual function, Top. SpinalCord Inj. Rehabil., 5, 21, 1999.

4. Kalra, L., The influence of stroke unit rehabilitation on functional recoveryfrom stroke, Stroke, 25, 821, 1994.

5. Chae, J. et al., Neuromuscular stimulation for upper extremity motor andfunctional recovery in acute hemiplegia, Stroke, 29, 975, 1998.

6. McColl, M. and Bickenbach, J., Introduction to Disability, W.B. Saunders, Lon-don, 1998.

7. Dzienkowski, R. et al., Cerebral palsy: a compressive review, Nurse Practitio-ner, 45, 1996.

8. Triolo, R., Kobetic, R., and Betz, R., Standing and walking with FNS: technicaland clinical challenges, in Human Motion Analysis: Current Applications andFuture Directions, Harris, G. and Smith P., Eds., IEEE Press, New York, 1996,p. 318.



9. Benton, L.A. et al., Functional Electrical Stimulation: A Practical Guide, 2nd ed.,Rancho Los Amigos Medical Center, California, 1981.

10. Edelberg, R., Electrical properties of the skin, in Biophysical Properties of theSkin, Elden, H.R., Ed., Wiley, New York, 1971, p. 513.

11. Memberg, W.D. et al., An analysis of the reliability of percutaneous intramus-cular electrodes in upper extremity FNS applications, IEEE Trans. Rehabil.Eng., 1, 126, 1993.

12. Scheiner, A., Polando, G., and Marsolais, E., Design and clinical applicationof a double helix electrode for functional electrical stimulation, IEEE Trans.Biomed. Eng., 41, 425, 1984.

13. Gradjean, P.A. and Mortimer, J.T., Recruitment properties of monopolar andbipolar epimysial electrodes, Ann. Biomed. Eng., 14, 53, 1986.

14. Memberg, W., Peckham, P., and Keith, M., A surgically implanted intramus-cular electrode for an implantable neuromuscular stimulation system, IEEETrans. Rehabil. Eng., 2, 80, 1994.

15. Naples, G. et al. A spiral nerve cuff electrode for peripheral nerve stimulation,IEEE Trans. Biomed. Eng., 35, 905, 1988.

16. McNeal, D.R. and Bowman, B.R., Selective activation of muscles using pe-ripheral nerve electrodes, Med. Biol. Eng. Comput., 23, 249, 1985.

17. Barone, F.C. et al., A bipolar electrode for peripheral nerve stimulation, BrainRes. Bull., 4, 421, 1979.

18. Veraart, C., Grill, W.M., and Mortimer, J.T., Selective control of muscle acti-vation with a multipolar nerve cuff electrode, IEEE Trans. Biomed. Eng., 40,640, 1993.

19. Scott, T., Peckham, P.H., and Kilgore, K., Tri-state myoelectric control of bi-lateral upper extremity neuroprostheses for tetraplegic individuals, IEEETrans. Rehabil. Eng., 4, 251, 1996.

20. Peckham, P., Mortimer, J., and Marsolais, E., Controlled prehension and re-lease in the C5 quadriplegic elicited by functional electrical stimulation of theparalyzed forearm musculature, Ann. Biomed. Eng., 8, 369, 1980.

21. Saxena, A., Nikolic, S., and Popovic, D., An EMG-controlled grasping systemfor tetraplegics, J. Rehabil. Res. Dev., 32, 17, 1995.

22. Hart, R., Kilgore, K., Peckham, P.H., A comparison between control methodsfor implanted FES hand-grasp systems, IEEE Trans. Rehabil. Eng., 6, 1, 1998.

23. Johnson, M. and Peckham, P.H., Evaluation of shoulder movement as a com-mand control source, IEEE Trans. Biomed. Eng., 37, 876, 1990.

24. Johnson, M. et al., Implantable transducer for two-degree-of-freedom jointangle sensing, IEEE Trans. Rehabil. Eng., 7, 349, 1999.

25. Nathan, R., Control strategies in FNS systems for the upper extremities, Crit.Rev. Biomed. Eng., 21, 485, 1993.

26. Graupe, D. and Kohn, K., Transcutaneous functional electrical stimulation ofcertain traumatic complete thoracic paraplegics for independent short-dis-tance ambulation, Neurol. Res., 19, 323, 1997.

27. Hoshimiya, N. et al., A multichannel FES system for the restoration of motorfunctions in high spinal cord injury patients: a respiration-controlled systemfor multijoint upper extremity, IEEE Trans. Biomed. Eng., 36, 754, 1989.

28. Handa, Y. et al., A voice controlled functional electrical stimulation systemfor the paralyzed hand, Jpn. J. Med. Electron. Biol. Eng., 23, 292, 1985.

29. Handa, Y. and Hoshimiya, N., Functional electrical stimulation for the controlof the upper extremities, Med. Prog. Technol, 12, 51, 1987.



30. Nathan, R. and Ohry, A., Upper limb functions regained in quadriplegia: ahybrid computerized FNS system, Arch. Phys. Med. Rehabil., 71, 415, 1990.

31. Lauer, R., Peckham, P.H., and Kilgore, K., EEG-based control of a hand graspneuroprosthesis, NeuroReport, 8, 1767, 1999.

32. Scott, T., Peckham, P.H., and Keith, M., Upper extremity neuroprosthesesusing functional electrical stimulation, in Clinical Neurology: International Prac-tice and Research, Brindley, G. and Rushton, D., Eds., Balliere Tindall, London,1995.

33. Lauer, R.T. et al., Applications of cortical signals to neuroprosthetic control:a critical review, IEEE Trans. Rehabil. Eng., 8, 205, 2000.

34. Mortimer, J. et al., Shoulder position transduction for proportional two axiscontrol of orthotic/prosthetic system, in The Control of Upper-Extremity Pros-theses and Orthoses, Herberts, P. et al., Eds., Charles C Thomas, Springfield,IL, 1974.

35. Shannon, C. and Weaver, W., The Mathematical Theory of Communication, TheUniversity of Illinois Press, Urbana, 1949.

36. Schwartz, M., Information Transmission, Modulation and Noise, McGraw-Hill,New York, 1980.

37. Schmidt, E. et al., Fine control of operantly conditioned firing patterns ofcortical neurons, Exper. Neurol., 61, 349, 1978.

38. Meek, S. and Fetherston, S., Comparison of signal-to-noise ratio of myoelectricfilters for prosthesis control, J. Rehabil. Res. Dev., 29, 9, 1992.

39. Welford, A., Fundamentals of Skill, Butler & Tanner, London, 1971.40. Doubler, J. and Childress, D., An analysis of extended physiological propri-

oception as a prosthesis-control technique, J. Rehabil. Res. Dev., 21, 5, 1984.41. Doubler, J. and Childress, D., Design and evaluation of a prosthetic control

system based on the concept of extended physiological proprioception, J.Rehabil. Res. Dev., 21, 19, 1984.

42. Guyton, A., Textbook of Medical Physiology, 8th ed., W.B. Saunders, Philadel-phia, 1991, chap. 22.

43. Neuman, M.R., Therapeutic and prosthetic devices, in Medical Instrumentation:Application and Design, 2nd ed., Webster, J., Ed., Houghton-Mifflin, Boston,1992, chap. 13.

44. Roy, O.Z., The current status of cardiac pacing, CRC Crit. Rev. Bioeng., 2, 259,1975.

45. Harthorne, J.W., Pacemaker leads, Int. J. Cardiol., 6, 423, 1984.46. Furman, S. and Escher, D., Principles and Techniques of Cardiac Pacing, Harper

& Row, New York, 1970.47. Liberson, W.T. et al., Functional electrotherapy: stimulation of the peroneal

nerve synchronized with the swing phase of the gait of hemiplegic patients,Arch. Phys. Med. Rehabil., 42, 101, 1961.

48. Long, C. and Masciarelli, V., An electrophysiological splint for the hand, Arch.Phys. Med. Rehabil., 44, 449, 1963.

49. Mortimer, J., Motor prostheses, in Handbook of Physiology: The Nervous SystemII, Brookhart, J.M. and Mountcastle, V.B., Eds., American Physiological Soci-ety, Bethesda, MD, 1981, p. 155.

50. Akers, J.M. et al., Tissue response to chronically stimulated implanted epimy-sial and intramuscular electrodes, IEEE Trans. Rehabil. Eng., 5, 207, 1997.



51. van den Honert, C. and Mortimer, J.T., Generation of unidirectionally prop-agated action potentials in a peripheral nerve by brief stimuli, Science, 206,1311, 1979.

52. Chae, J. et al., Functional neuromuscular stimulation, in Rehabilitation Medi-cine: Principles and Practice, 3rd ed., DeLisa, J.A. and Gans, B.M., Eds., Lippin-cott-Raven, Philadelpiha, 1998.

53. Zafar, M. and Van Doren, C., Effectiveness of supplemental grasp-force feed-back in the presence of vision, Med. Biol Eng. Comput., 38, 267, 2000.

54. Johnson, K.O. et al., Perspectives on the role of afferent signals in control ofmotor neuroprostheses, Med. Eng. Phys., 17, 481, 1995.

55. Bhadra, N., Kilgore, K.L., and Peckham, P.H., Implanted stimulators for res-toration of function in spinal cord injury, Med. Eng. Phys., 23, 19, 2001.

56. Inmann, A. et al., Signals from skin mechanoreceptors used in control of ahand grasp neuroprosthesis, NeuroReport, 12, 2817, 2001.

57. Haugland, M. and Sinkjaer, T., Interfacing the body’s own sensing receptorsinto neural prosthesis devices, Technol. Health Care, 7, 393, 1999.

58. Cameron, T. et al., The effect of wrist angle on electrically evoked handopening in patients with spastic hemiplegia, IEEE Trans. Rehabil. Eng., 7, 109,1999.

59. Prochazka, A. et al., The bionic glove: an electrical stimulator garment thatprovides controlled grasp and hand opening in quadriplegia, Arch. Phys. Med.Rehabil., 78, 608, 1997.

60. Weingarden, H.P. et al., Hybrid functional electrical stimulation orthosis sys-tem for the upper limb: effects on spasticity in chronic stable hemiplegia, Am.J. Phys. Med. Rehabil., 77, 276, 1998.

61. Weingarden, H.P. et al., Upper limb functional electrical stimulation for walk-er ambulation in hemiplegia: a case report, Am. J. Phys. Med. Rehabil., 76, 63,1997.

62. Triolo, R. et al., Challenges to clinical deployment of upper limb neuropros-theses, J. Rehabil. Res. Dev., 33, 111, 1996.

63. Kraft, G.H., Fitts, S.S., and Hammond, M.C., Techniques to improve functionof the arm and hand in chronic hemiplegia, Arch. Phys. Med. Rehabil., 73, 220,1992.

64. Peckham, P.H. et al., Efficacy of an implanted neuroprosthesis for restoringhand grasp in tetraplegia: a multicenter study, Arch. Phys. Med. Rehabil., 82,1380, 2001.

65. Stroh-Wuolle, K. et al., Satisfaction with and usage of a hand neuroprosthesis,Arch. Phys. Med. Rehabil., 80, 206, 1999.

66. Smith, B. et al., An externally powered, multichannel, implantable stimulator-telemeter for control of paralyzed muscle, IEEE Trans. Biomed. Eng., 45, 463,1998.

67. Kilgore, K.L. et al., An implanted upper-extremity neuroprosthesis. Follow-up of five patients, J. Bone Joint Surg. Am., 79, 533, 1997.

68. Wijman, C.A. et al., Functional evaluation of quadriplegic patients using ahand neuroprosthesis, Arch. Phys. Med. Rehabil., 71, 1053, 1990.

69. Stein, R.B., Functional electrical stimulation after spinal cord injury, J. Neu-rotrauma, 16, 713, 1999.

70. Wieler, M. et al., Multicenter evaluation of electrical stimulation systems forwalking, Arch. Phys. Med. Rehabil., 80, 495, 1999.



71. Taylor, P. N et al., Clinical use of the Odstock dropped foot stimulator: itseffect on the speed and effort of walking, Phys. Med. Rehabil., 80, 1577, 1999.

72. Taylor, P.N. et al., Patients’ perceptions of the Odstock dropped foot stimu-lator (ODFS), Clin. Rehabil., 13, 439, 1999.

73. Kralj, A., Acimovic, R., and Stanic, U., Enhancement of hemiplegic patientrehabilitation by means of functional electrical stimulation, Prosthet. Orthot.Int., 17, 107, 1993.

74. Kralj, A. and Bajd, T., Functional Electrical Stimulation: Standing and Walkingafter Spinal Cord Injury, CRC Press, Boca Raton, FL, 1989.

75. Grill, W.M. and Kirsch, R., Neuroprosthetic Applications of electrical stimu-lation, Asst. Technol., 12, 6, 2000.

76. Graupe, D. et al., Ambulation by traumatic T4–12 paraplegics using functionalneuromuscular stimulation, CRC Crit. Rev. Neurosurg., 8, 221, 1998.

77. Graupe, D. and Kohn, K.H., Functional neuromuscular stimulator for short-distance ambulation by certain thoracic-level spinal-cord-injured paraplegics,Surg. Neurol., 50, 202, 1998.

78. Popovic, M.R. et al., Functional electrical stimulation for grasping and walk-ing: indications and limitations, Spinal Cord, 39, 403, 2001.

79. Creasey, G.H. et al., An implantable neuroprosthesis for restoring bladderand bowel control to patients with spinal cord injuries: a multicenter trial,Arch. Phys. Med. Rehabil., 82, 1512, 2001.

80. Brindley, G.S., The first 500 patients with sacral anterior root stimulator im-plants: general description, Paraplegia, 32, 795, 1994.

81. Linder, S.H., Functional electrical stimulation to enhance cough in quadriple-gia, Chest, 103, 166, 1993.

82. Bryden, A.M., Memberg, W.D., and Crago, P.E., Electrically stimulated elbowextension in persons with C5/C6 tetraplegia: a functional and evaluation,Phys. Med. Rehabil., 81, 80, 2000.

83. Lauer, R.T. et al., The function of the finger intrinsic muscles in response toelectrical stimulation, IEEE Trans. Rehabil. Eng., 7, 19, 1999.

84. Kameyama, J. et al., Electromyographic study relating to shoulder motion:control of shoulder joint by functional electrical stimulation, Tohoku J. Exp.Med., 187, 339, 1999.

85. Kameyama, J. et al., Restoration of shoulder movement in quadriplegic andhemiplegic patients by functional electrical stimulation using percutaneousmultiple electrodes, Tohoku J. Exp. Med., 187, 329, 1999.

86. Handa, Y., Yagi, R., and Hoshimiya, N., Application of functional electricalstimulation to the paralyzed extremities, Neurol. Med. Chir. (Tokyo), 38, 784,1998.

87. Ichie, M. et al., Control of thumb movements: EMG analysis of the thumband its application to functional electrical stimulation for a paralyzed hand,Frontiers Med. Biol. Eng., 6, 291, 1995.

88. Handa, Y. et al., Functional electrical stimulation (FES) systems for restorationof motor function of paralyzed muscles — versatile systems and a portablesystem, Frontiers Med. Biol Eng., 4, 241, 1992.

89. Popovic, M.R. et al., Surface-stimulation technology for grasping and walkingneuroprosthesis, IEEE Eng. Med. Biol. Mag., 20, 82, 2001.

90. Mackenzie-Knapp, M., Electrical stimulation in early stroke rehabilitation ofthe upper limb with inattention, Aust. J. Physiother., 45, 223, 1999.



91. Chae, J. and Yu, D., A critical review of neuromuscular electrical stimulationfor treatment of motor dysfunction in hemiplegia, Assist. Technol., 12, 33, 2000.

92. Dimitrijevic, M. and Soroker, N., Mesh-glove. 2: Modulation of residual upperlimb control after stroke with whole hand electrical stimulation, Scand. J.Rehabil. Med., 26, 187, 1994.

93. Francisco, G. et al., Electromyogram-triggered neuromuscular stimulation forimproving the arm function of acute stroke survivors: a randomized pilotstudy, Arch. Phys. Med. Rehabil., 79, 549, 1996.

94. Carmick, J., Use of neuromuscular electrical stimulation and dorsal wristsplint to improve the hand function of a child with spastic hemiparesis, Phys.Ther., 77, 661, 1997.

95. Wood, D.E. et al., Is paraplegic standing by root stimulation a practical option?Conclusions from the LARSI project, in Proc. 6th Annual Conference of theIFESS, Cleveland, OH, 2001, p. 13.

96. Davis, R., MacFarland, W., and Emmons, S., Initial results of the Nucleus FES-22 implanted stimulator for limb movement in paraplegia, Stereotact. Funct.Neurosurg., 63, 192, 1994.

97. Davis, R., Patrick, J., and Barriskill, A., Development of functional electricalstimulators utilizing cochlear implant technology, Med. Eng. Phys., 23, 61,2001.

98. Guiraud, D. et al., One year implanted patients follow up: SUAW project firstresults, in Proc. 6th Annual Conference of the IFESS, Cleveland, OH, 2001, p. 55.

99. Davis, J.A. et al., Preliminary performance of a surgically implanted neuro-prosthesis for standing and transfers — where do we stand?, J. Rehabil. Res.Dev., 38, 609, 2001.

100. Davis, J.A. et al., Surgical technique for installing an eight-channel neuropros-thesis for standing, Clin. Orthop., 385, 237, 2001.

101. Uhlir, J.P., Triolo, R.J., and Kobetic, R., The use of selective electrical stimula-tion of the quadriceps to improve standing function in paraplegia, IEEE Trans.Rehabil. Eng., 8, 514, 2000.

102. Haugland, M. and Sinkjaer, T., Interfacing the body’s own sensing receptorsinto neural prosthesis devices, Technol. Health Care, 7, 393, 1999.

103. Bosnjak, R., Dolenc, V., and Kralj, A., Biomechanical response in the ankle tostimulation of lumbosacral nerve roots with spiral cuff multielectrode —preliminary study, Neurol. Med. Chir. (Tokyo), 39, 659, 1999.

104. Bajd, T. et al., Use of functional electrical stimulation in the lower extremitiesof incomplete spinal cord injured patients, Artif. Organs, 23, 403, 1999.

105. Kralj, A. et al., FES gait restoration and balance control in spinal cord-injuredpatients, Prog. Brain Res., 97, 387, 1993.

106. Daly, J.J. et al., Feasibility of gait training for acute stroke patients using FNSwith implanted electrodes, J. Neurol. Sci., 179, 103, 2000.

107. Daly, J.J. and Ruff, R.L., Electrically induced recovery of gait components forolder patients with chronic stroke, Am. J. Phys. Med. Rehabil., 79, 349, 2000.

108. Pearson, J.A. and Wenkstern, B., Habituation and sensitization of the flexorwithdrawal reflex, Brain Res., 43, 107, 1972.

109. Grill, W.M. et al., Emerging clinical applications of electrical stimulation:opportunities for restoration of function, J. Rehabil. Res. Dev., 38, 641, 2001.

110. Bidus, K.A., Thomas, G.R., and Ludlow, C.L., Effects of adductor musclestimulation on speech in abductor spasmodic dysphonia, Laryngoscope, 110,1943, 2000.



111. Barkmeier, J.M. et al., Modulation of laryngeal responses to superior laryngealnerve stimulation by volitional swallowing on awake humans, J. Neurophys-iol., 83, 1264, 2000.

112. Curbelo, R. et al., LD Pace II, an easily programmable device for cardiomy-oplasty, Med. Eng. Phys., 23, 45, 2001.

113. Hogan, J.F., Koda, H., and Glenn, W.W., Electrical techniques for stimulationof the phrenic nerve to pace the diaphragm: inductive coupling and batterypowered total implant in asynchronous and demand modes, Pacing Clin.Electrophysiol., 12, 847, 1989.

114. Glenn, W.W. et al., Fundamental considerations in pacing of the diaphragmfor chronic ventilatory insufficiency: a multi-center study, Pacing Clin. Elec-trophysiol., 11, 2121, 1988.

115. Sauermann, S. et al., Computer aided adjustment of the phrenic pacemaker:automatic functions, documentation, and quality control, Artif. Organs, 21,216, 1997.

116. Girsch, W. et al., Vienna phrenic pacemaker — experience with diaphragmpacing in children, Eur. J. Pediatr. Surg., 6, 140, 1996.

117. DiMarco, A.F. et al., Evaluation of intercostal pacing to provide artificialventilation in quadriplegics, Am. J. Respir. Crit. Care Med., 163, A151, 2001.

118. DiMarco, A.F., Neural prostheses in the respiratory system, J. Rehabil. Res.Dev., 38, 601, 2001.

119. Kirkham, A.P. et al., The acute effects of continuous and conditional neuro-modulation on the bladder in spinal cord injury, Spinal Cord, 39, 420, 2001.

120. Craggs, M.D. et al., SPARSI: an implant to empty the bladder and controlincontinence without posterior rhizotomy in spinal cord injury, Brit. J. Urol.Int., 85, 2, 2000.

121. Grill, W.M., Bhadra, N., and Wang, B., Bladder and urethral pressures evokedby microstimulation of the sacral spinal cord in cats, Brain Res., 836, 19, 1999.

122. Grill, W.M. et al., At the interface: convergence of neural regeneration andneural prostheses for restoration of function, J. Rehabil. Res. Dev., 38, 633, 2001.

123. Borgens, R.B., Electrically mediated regeneration and guidance of adult mam-malian spinal axons into polymeric channels, Neuroscience, 91, 251, 1999.

124. Borgens, R.B., Blight, A.R., and McGinnis, M.E., Functional recovery afterspinal cord hemisection in guinea pigs: the effects of applied electrical fields,J. Comp. Neurol., 296, 634, 1990.

125. Loeb, G.E. et al., BION system for distributed neural prosthetic interfaces,Med. Eng. Phys., 23, 9, 2001.

126. Cameron, T. et al., Micromodular implants to provide electrical stimulationof paralyzed muscles and limbs, IEEE Trans. Biomed. Eng., 44, 781, 1997.

127. Dupont, A. et al., Clinical trials of BION injectible neuromuscular stimulators,in Proc. 6th Annual Conference of the IFESS, Cleveland, OH, 2001, p. 7.

128. Chen, T. et al., A miniature biofuel cell, Am. Chem. Soc., 123, 8630, 2001.129. Chapin, J. et al., Real-time control of a robot arm using simultaneously re-

corded neurons in the motor cortex, Nat. Neurosci., 2, 664, 1999.130. Maynard, E. et al., Neuronal interactions improve cortical population coding

of movement direction, J. Neurosci., 19, 8083, 1999.131. Schwartz, A., Motor cortical activity during drawing movements: population

representation during sinusoid tracing, J. Neurophysiol., 70, 28, 1992.




section six

Emerging technologies


chapter thirteen

Neurotechnology: microelectronics

Danny Banks

Contents

13.1 Introduction13.1.1 The need for integration13.1.2 Noise sources in recording13.1.3 MEMS neuroprosthetic system outlines

13.1.3.1 CNS applications13.1.3.2 PNS applications13.1.3.3 The eye

13.1.4 Power dissipation in integrated circuits13.1.5 MOS or bipolar ?

13.1.5.1 Semiconductor basics13.1.5.2 Bipolar13.1.5.3 The MOSFET13.1.5.4 The JFET

13.2 Elements of a system13.2.1 The neuron–transistor junction13.2.2 General

13.2.2.1 Clocks and oscillators13.2.2.2 Power-on reset and mode control13.2.2.3 Integrators13.2.2.4 Phase-locked loops

13.2.3 Recording13.2.3.1 Signal preamplifiers13.2.3.2 Filtering13.2.3.3 Analog-to-digital converters



13.2.4 Stimulation13.2.4.1 Voltage-output DACs13.2.4.2 Current output DACs

13.2.5 Elements specific to CNS prostheses13.2.6 Elements specific to PNS prostheses

13.2.6.1 Voltage regulation13.2.6.2 Communication schemes

13.2.7 Elements specific to retina prostheses13.2.8 Other elements

13.3 Future directionsAcknowledgmentsReferences

13.1 IntroductionIn recent decades, the development of neural prostheses has benefitedgreatly from technological advances, particularly in the area of greaterminiaturization and integration of microelectronics. The development ofglass microelectrodes early in the 20th century and the later developmentof metal wire microelectrodes in the 1950s1 gave the neurophysiologist newinsights into the operation of the nervous system. As early as the 1960s, itwas being proposed that integrated circuit (IC) fabrication techniquescould be employed to develop improved devices. By the early 1970s,devices had been fabricated and a design had been published for animplantable monolithic wafer electrode.2,3 In 1975, Wise and Angell pub-lished an improved version4 of their 1970 design, which included transistorbuffer amplifiers.

Since this time, microelectromechanical systems (MEMS) techniqueshave improved markedly and have been used to develop a variety of dif-ferent microelectrode structures. These techniques are reviewed in otherchapters of this volume. The problems relating to the integration of micro-electronic circuitry onto these MEMS structures for neuroprosthetic applica-tions are often regarded as problems that will be overcome with the continualreduction of feature size (driven by demand from the microelectronics indus-try) and complementary metal-oxide semiconductor (CMOS)-compatibleMEMS processes (driven by demand from the MEMS industry). Nonetheless,the integration of MEMS technologies with microelectronics for neuropros-thetic applications presents particular problems. It is the intention of thischapter to present the reader with an overview of the available microelec-tronics technologies and approaches that have been used to integrate thesewith MEMS-based microelectrode structures. This is done within the contextof some generic “MEMS neuroprosthetic systems,” outlined below. First,however, it is necessary to make a case for increasing integration in neuro-prosthetic applications.



13.1.1 The need for integration

The use of MEMS-based microelectrode structures offers clear advantages.They allow more precise stimulation with lower stimulation current require-ments than would otherwise be obtainable with larger devices; this wasrecognized at an early stage in attempts to develop a prosthesis that stimu-lated the visual cortex.5 Large numbers of electrode sites provide for redun-dancy and the possibility to compensate for drift of individual neurons inand out of range of the recording or stimulation electrode.6,7 Miniaturizationalso facilitates the placement of large numbers of electrode sites in smallvolumes of relatively inaccessible tissue, which provides advantages overother approaches.8

In and of itself, miniaturization of the electrode structure alone is notan ideal solution. Neuroprostheses have suffered, and may still suffer, fromreliability problems when a large amount of wiring is used to interconnectthe many different components involved: cuff electrodes, controller, etc.By introducing an even larger number of electrodes, through miniaturiza-tion, these problems are compounded. Additionally, the MEMS structureitself is at a disadvantage when connected by a large number of cables.MEMS-based microelectrodes have often been designed to be floatingstructures, so that they are able to move with the nervous system. Attachingwires to the device has a tethering effect. Not only are the connectionsliable to break through mechanical stress, but the device is now more likelyto move relative to the neurons that it is stimulating/recording from,causing unwanted tissue damage. This is of great concern in retinal pros-theses where the mechanical properties of the retina have been likened tothose of wet tissue paper.

Furthermore, the use of MEMS-based devices for recording in neuropy-siological investigations where implantation and mobility of the specimenhave not been of such concern9 has revealed an additional problem — thatof dealing with the sheer volume of data generated by systems incorporatinglarge numbers of microelectrode recording sites capable of recording detailsof individual action potential spikes. This is of particular concern inimplanted neuroprosthetic systems where real-time control is imperative yetcomputing power, as well as electrical power to supply the computing ele-ment, is limited.

Thus, the need for intelligent MEMS-based microelectrode devices hasbeen established that offer reduction or elimination of interconnecting leadsand distribution of computing power. It then becomes necessary to considerthe approach to be taken to achieve this desirable goal. The details of thisare considered below, but the researcher/designer is faced with one initialdecision to make: whether or not to select a monolithic approach, in whichthe microelectronics is integrated into the same substrate from which themicroelectrodes are machined, or a hybrid approach, where the electrodesand microelectronic circuitry are fabricated using different and often incom-patible technologies, and then later assembled into one unit.



The hybrid approach enables the designer to select the most appropriatetechnology for each component. The Fraunhöfer Institute has, for example,achieved some remarkable results with regeneration electrode devices fab-ricated on polymer substrates.10,11 However, technology for fabricating elec-tronic components using polymers is in its infancy. Therefore, signal pro-cessing circuitry is fabricated on a silicon chip that is bonded to the polymericelectrode structure.12

There are, however, several disadvantages to the hybrid approach. Thefirst is that of cost. Once the batch production regime of microfabricationhas been abandoned and components have to be handled individually, costsincrease dramatically. For research purposes, however, this is not of such agreat concern as the economies of scale are not there, and one will often wishto examine different combinations of components as potential solutions toa problem. Of greater concern is that of reliability. Each additional compo-nent that must be assembled into a system increases the chance of systemfailure, and each additional bonded joint is a potential point of failure.

The disadvantage of the monolithic approach is that a compromise tech-nology has to be developed, and this may not necessarily lead to the bestpossible system performance. Monolithic solutions will generally take longerto develop than hybrid solutions, and the initial expenditure on researchwill be greater; they will generally lag hybrid approaches in performancebut should eventually yield less expensive and more reliable solutions.

It is now possible to summarize the reasons for, goals, and require-ments of greater integration of microelectronics with MEMS-based elec-trode structures:

• Reduce or eliminate interconnecting leads.• Distribute intelligence.• Minimize power consumption.• Minimize device area (volume).• Improve reliability and longevity of implants.

Device area is usually referred to rather than volume because microelectroniccircuits are usually fabricated on one side of a silicon wafer and have neg-ligible height in comparison to the other two dimensions. Power consump-tion must be minimized not only from the point of view of the difficulty oftransmitting electrical power across the skin, but also from the point of viewof power dissipated from the implanted device into the surrounding tissue.Najafi and Wise13,14 provide an example of this consideration.

13.1.2 Noise sources in recording

The neuroprosthesis will consist of either recording electrodes or stimulationelectrodes, or both. When considering stimulation, the circuit design prob-lems that arise are concerned with selecting the electrode combination to beused, efficiently delivering a stimulating current, controlling the shape and



charge balance of the stimulating pulse, and ensuring that the device hassufficient power to perform the required stimulation. These circuit designconsiderations are dealt with in further detail below. Recording differs fromstimulation in that, while the latter involves the delivery of relatively largesignals into a resistive/reactive load, the former involves trying to detectrelatively small voltage signals through a large source impedance. Beforediscussing the circuitry that has been proposed to solve the problems thatthis gives rise to, it is useful to lay out the problems to be solved by therecording circuitry in more detail.

The noise sources encountered when recording can be classified as thoseintrinsic to the electrode design, over which the designer of the microelec-tronic circuitry has no control, and those over which the designer of themicroelectronic circuitry has control. In the first camp, one has the sourceimpedance of the electrode-tissue interface and the spreading resistance ofthe microelectrode recording site. In his analysis of noise sources related toelectrode design, Edell15 concluded that the spreading resistance of the elec-trode site was what would provide a physical limit to recording site size andhence selectivity. One other noise source over which the designer of themicroelectronics has no control is the slow base-line drift, or even the half-cell potential, of the recording-site tissue interface.13 Those aspects overwhich the designer of the microelectronic circuitry has considerable controlare those relating to noise in the signal-processing circuitry itself, electro-magnetic pick-up (or interference [EMI]), and stimulation artifacts. One mustalso consider the problem of bias currents passing across the microelec-trode–tissue interface.

Consideration of signal processing, or amplifier noise, and EMI providesadditional support for integrating amplifiers as close as possible to the micro-electrode recording sites. EM pick-up is caused by induction of currents froman external source of EMI (e.g., radiofrequency sources, although mainselectricity supply lines are also problematic). The normal solution to this isto employ a differential amplifier with a high common-mode rejection ratio(CMRR); this is a measure of how well the amplifier responds when thesame signal appears on both inputs (for a differential amplifier, the outputshould be zero) and is defined as the ratio of the gain in difference mode tothe gain in common mode (the same signal applied to both inputs; seeBogart24). This, of course, requires that the signals induced in the two leadsof the amplifier by the same EMI source be as similar as possible. EMI ispartially related to the area of any circuit loops into which currents areinduced, so normally an effort is made to minimize the area of any loops inthe circuit; however, this is usually more applicable to printed circuit board(PCB) design than integrated circuit (IC) design.

Capacitive coupling between the tracks that connect the recording sites tothe inputs of the amplifiers is also a potential source of interference (cross-talk),where a large signal on one recording site appears as an artifact in the recordof an adjacent site. Najafi et al.16 have analyzed how this affects electrode designand scaling for silicon-based devices, concluding that cross-talk should not be



expected to be a limiting factor for purely recording devices, and data arepresented to assist in the design of mixed recording/stimulating devices.

The importance of having the highest possible gain on the first amplifi-cation stage, situated as electrically close as possible to the recording site,should also be mentioned when discussing noise sources. Consider twoamplification stages, the first with gain A1 and the second with gain A2. Thetotal voltage noise due to the first amplifier stage (referred to input) will beVN1, and the noise referred to input of the second stage will be VN2. Ideally,VN1 will be insignificant compared to the voltage noise level present in thesignal, but this may not always be achievable. We consider these signals tobe random processes and thus sum their squares. Thus, the total noisecontribution arising from the amplifiers will be VN = .

From this it is possible to see that any gain above unity provided by A1

will tend to make the VN1 term dominant, thus the noise arising from sub-sequent circuitry may be neglected, although the designer should not takethis as an allowance for bad design of subsequent analog stages. Note, forinstance, that noise appearing on the power rails, or through other mecha-nisms, due to poorly designed analog or digital stages can feed back anddegrade the performance of the initial amplification stage.

The question of interference from unwanted bioelectric sources, such asEMG, straddles the divide between microelectronics design and microelec-trode design. The use of insulating nerve cuffs is generally effective in exclud-ing or significantly reducing interference from extraneural sources. However,their effect of amplifying signals within the nerve trunk being recorded frommay well counteract a primary reason for using MEMS-based microelectrodedevices: increased recording selectivity.

Edell’s analysis has elegantly demonstrated that by recording differen-tially between two adjacent sites on the device, utilizing the common-moderejection characteristics of the amplifier, the magnitude of the signal dropsoff with the square of distance when the distance to the source becomes muchgreater than the distance separating the two recording sites.15 When the sourceis closer to the pair of recording sites, the magnitude of the signal would beinversely proportional to distance (approximately). This suggests that differ-ential recording between adjacent sites could be effectively employed toreduce noise and increase specificity of recording. Unfortunately, a simpledifferential configuration also has a “dead spot” directly between the tworecording sites. On a typical MEMS-based microelectrode structure, theserecording sites will either be on the same bar or shank, or on adjacent barsor shanks. In the first case, the recording sites will be most sensitive to thesources along the axis of the shank — those most likely to have been damagedby insertion or where connective tissue would have grown around a chronicimplant. In the second case, the healthiest tissue, from which one would wishto record, lies precisely in the dead spot between the two recording sites.Clearly, then, a simple differential configuration cannot be readily adopted.Alternative recording site/reference electrode layouts could be investigated

A12A2

2VN12 A2

2VN22+



for differential recording. This would, however, have an impact on the micro-electronics design. Generally, it would be desirable to situate the differentialstage as electrically close as possible to the electrode sites involved to benefitfrom the CMRR of the amplifiers. With good analog circuit design, however,the designer may prefer to use computing power to combine the signalsappearing at different electrode sites once they have been digitized. Onereason that this may be beneficial is that it would be a much more flexibleapproach, perhaps enabling the processor to track individual signal sourcesto some extent. Information lost due to the implementation of differentialrecording in hardware cannot be recovered again.

13.1.3 MEMS neuroprosthetic system outlines

From the literature, it is possible to generalize three different situations forMEMS-based neuroprosthetic systems and to identify similarities and dif-ferences between them that affect the circuit design. These are systemslocated in the central nervous system (CNS), systems located in the periph-eral nervous system (PNS), and those located in the eye (retinal prostheses).With this in mind, it is possible to outline generic systems, and use these asa basis for discussion of the individual circuit elements involved. Many ofthese circuit elements will overlap from one system to another, but there aresome unique problems to be solved for the different applications; particularlywhen one considers the retinal prosthesis.

13.1.3.1 CNS applicationsIn this generalization, prostheses applied to the brain are considered. Thespinal cord represents something of a transitional area, exhibiting some ofthe characteristics one may find relating to prosthetics applied to the brainand some of those relating to the PNS. Specific examples of such prosthesesinclude visual prosthetics applied to the visual cortex,5,17,18 prostheticsapplied to the auditory brain stem,19,20 and those applied to the motor cor-tex.21,22 It is useful to note that the latter represent, to some extent, an attemptto link CNS and PNS prosthetics in the case of spinal cord injury. It has beenproposed that lost sensation and control in paralyzed limbs can be restoredby recording from the motor cortex and interpreting and transmitting thesecommands to a PNS element of the prosthesis. This would, in turn, electri-cally stimulate the PNS/muscles to achieve the desired effect. Proprioceptivesignals recorded from the PNS would be used to control this stimulationand those returned to the CNS element of the prosthesis would provideconscious feedback of the process by stimulating the sensory cortex of thepatient.23 This is, perhaps, one of the most challenging systems so far pro-posed for MEMS-based neuroprosthetics. The principle characteristics of theenvironment encountered by prosthetics applied to the brain, from an elec-tronics point of view, are the relative mechanical stability, the potential ofmounting connectors on the skull, the relative lack of EMG and other inter-ference, and complexity.



We can, therefore, identify a generic CNS neuroprosthetic system con-sisting of a fairly substantial two- or three-dimensional electrode array, pen-etrating up to a few millimeters into the brain. Although the complexity ofbrain means that the signal processing and identification overhead involvedcan be relatively large, the location of the device in the relatively stable andshielded environs of the skull means that a relatively large area of siliconreal estate can be located relatively near to the electrode structure itself.Additionally, the skull provides a good anchor point for percutaneous con-nectors, which can be employed (e.g., work of Dobelle et al.18), althoughsome may still prefer the radiofrequency (RF) induction coil or other trans-cutaneous approaches. The length of wiring is of less concern when allelements of a system can be mounted in close proximity, but the number ofconnections required, particularly to percutaneous connectors, is still a mat-ter of interest. This is because of the pulsation of the brain with blood flowand the potential tethering effect of the connector or other elements of thesystem. The sort of generic system that we consider can be illustrated by theworks of Dobelle18 and Kennedy et al.22 (see Figure 13.1a).

13.1.3.2 PNS applicationsThe PNS differs from the CNS in that the complexity of the system decreasesas one considers more distal locations; there is even some evidence forfunctional organization within the nerve trunk itself. Nerve cuff solutionshave been applied to the PNS with considerable success in the past, but morecomplex systems for standing and walking have suffered from problems ofcable failure.25 Where such systems have employed more proximally locatedcuff electrodes (at the ventral roots, for example), the success of the systemhas been limited by the selectivity of the electrodes employed.26

It is noted that while the PNS has evolved to withstand many decades ofrepeated mechanical stress from joint flexion and muscle contraction, the sameis not true of metal wires. Locating prosthetic systems beyond the relativelyprotected confines of the spinal column can be problematic; even here, spaceis at a premium, implying a requirement for a miniaturized system.

The PNS prosthesis will typically employ two-dimensional regenerationelectrodes or probe-type devices,16,27,28 although at least one group has pro-posed that a three-dimensional array would be required to interface in ahighly selective manner to individual nodes of Ranvier.29 One of the appli-cations typically proposed for such a system is standing and walking inparaplegics. Here, it is proposed, the highly selective nature of the MEMS-based microelectrode structures can be drawn upon. The system could belocated in or near the spinal column with electrodes situated in the ventralroots for stimulation and devices situated in the dorsal roots providingsensory feedback for closed-loop control.6,30 Another proposed applicationis to provide control and sensation for prosthetic limbs.10,27,28

The generic PNS prosthesis therefore consists of a number of microelec-trode structures at several discrete locations within the PNS. These will typ-ically perform minimal signal processing, due to the difficulty of supplying



such locations with cabling for power and signal transmission. Ideally, suchdevices would not project from the nerve trunks into which they had beenimplanted, due to mechanical considerations, and wireless communicationand power would be employed. Control would be centralized, typically in aunit implanted beneath the skin with a coil for RF coupling; this latter itemis not dealt with in this chapter (see Figure 13.1b). Injectable stimulators31,32

provide a well-developed example of this approach.

13.1.3.3 The eyeThe retinal prosthesis has received much attention recently,33–37 and a typ-ical system is outlined in Figure 13.1c. In such a system, an electrode array

Figure 13.1 (a) CNS (brain) prosthesis, characterized by co-location of recording andstimulating sites with control and signal processing circuitry and a relatively stablesituation for per- or transcutaneous signal transmission. (b) PNS prosthesis, charac-terized by distributed microelectrode devices with the requirement for more distrib-uted intelligence and localized power regulation; per- or transcutaneous power anddata transmission is still required. (c) The retina prosthesis. For the purposes of thiswork, the spinal cord lies somewhere between the CNS and PNS outline.



designed to stimulate the retinal ganglion cells is implanted either on topof or beneath the retina. Power and communications are provided by anoptical link either to the array itself or to a control circuit located withinthe eye away from the array but connected by cables. The retinal prosthesisis dealt with in some detail in Chapter 11 in this volume, and it possessessome unique characteristics when it comes to the integration of microelec-tronic circuitry.

Notably, a wireless approach appears to be the most practical whentransmitting power and data to the array. This implies that any circuitryshould consume as little power as possible, and the additional complicationof localized heating caused by power dissipated by the electronics also mustbe looked at anew. Additionally, the delicate nature of the retina means thatany implant must be minimally invasive and mechanically compatible (i.e.,soft, flexible, and shaped to the curvature of the eye) to minimize the damagecaused. This tends to point to the mutually incompatible goals of a polymericelectrode structure and one with integrated electronics.

13.1.4 Power dissipation in integrated circuits

The power available for an implanted system will invariably be limited unlessa percutaneous connector is used, in which case the system will be relativelygenerously supplied. Even so, power conservation is a question that needs tobe addressed. Power may be dissipated in one of several ways:38,39

• Static power dissipation• Dynamic (switching) power dissipation• Short circuit power dissipation• Leakage power

Power is dissipated whenever current flows through a device that has avoltage drop across it, given by the equation:

Power = I V

This static power dissipation is most prominent in analog circuits, whereresistors and bias chains are part of the circuit design. Analog circuits nor-mally require that transistors be biased into a partially “on” state (conductingcurrent, with a voltage drop across them) in order to fulfill their function.This is best combated by care in circuit design, ensuring that voltages andquiescent currents are as low as possible.

Dynamic, or switching, power dissipation is the principle mode of cur-rent loss in CMOS digital logic circuits. An ideal switch would go instanta-neously from an open circuit condition (maximum voltage dropped acrossit, but no current flowing) to short circuit (no voltage dropped across it, butmaximum current flowing). In such a case, no power would be dissipated,as either V or I would be zero. However, switching involves the charging



and discharging of capacitances within the circuit. This means that there isa period in which the voltage across a switch is falling (or rising) while thecurrent is rising (or falling), leading to power dissipation in the device. Thepower dissipated by a circuit switching at a frequency f with a capacitancec is given by:

Power = c f v2

Thus, losses can be reduced by reducing parasitic capacitances and by reduc-ing the clock frequencies used, but more significantly they can be reducedby reducing the supply voltage at which ICs operate.

In many circuit configurations, particularly digital logic, complementarytransistors appear across the power rails (see Figure 13.6, for example). Dif-ferent threshold voltages and operating parameters mean it is possible whenswitching that both transistors could be on at the same time. This momentaryshort circuit condition dissipates power (short circuit dissipation) and isrelated to the switching frequency as well as the technology employed.

Leakage current power dissipation occurs when subthreshold leakagecurrents prevent transistors from turning completely off. Until recently, thishas not been a significant problem in ICs, but with ever-reducing featuresizes it may become an additional consideration in circuit design.

13.1.5 MOS or bipolar?

The designer is faced with a number of possibilities when considering thedesign of integrated circuitry for MEMS-based neuroprostheses: a bipolarapproach, MOSFET, JFET, or a mixed approach (usually bi-CMOS). It is nowbecoming possible to integrate all these technologies with MEMS, althoughCMOS is currently favored. Within this chapter, the focus is almost entirelyon CMOS circuitry, however it is necessary at this point to introduce anddismiss other approaches. It is also intended to give the reader a very briefintroduction to transistors.

13.1.5.1 Semiconductor basicsThe reader is referred to introductory books on microelectronic circuit designand fabrication for a more complete treatment of this topic.24,39–42 Semicon-ductor devices are commonly fabricated from silicon into which differentimpurities have been introduced to alter its electrical characteristics. For thepurposes of this section, we consider n-type and p-type silicon. A diode isformed by the interface between a region of n-type and p-type silicon andallows electrical current to flow in only one direction (from p to n).

When integrated circuits are fabricated, the impurities that will form thedevices are introduced into different areas of the silicon wafer. Silicon withintroduced impurities is termed doped. Conducting layers, separated by insu-lating layers, are then built up on top of the devices thus formed, to inter-connect them in a meaningful manner. Typically these will be one or two



layers of polysilicon (this is silicon, but with a different structure than thatof the silicon wafer), and two or more (up to six or more) layers of metal,usually aluminum. The circuitry is usually no more than a few microns thickand is located on the surface of a wafer typically half a millimeter thick.Thus, it may be considered to be more or less two dimensional. Devices areusually depicted in cross section when describing their function but aretypically depicted in plan view when laying out a design. Figure 13.2 illus-trates a diode to familiarize the reader with this type of depiction.

13.1.5.2 BipolarThe bipolar junction transistor (BJT) consists of a thin base region, sand-wiched between the emitter and collector regions. These may be doped eithern–p–n or p–n–p (collector–base–emitter), leading to a device with two diodejunctions. During operation, a current applied to the base is used to controlthe larger current that flows between collector and emitter. When construct-ing analog circuits (e.g., amplifiers and filters), bipolar technology possessesthe advantages that it can achieve higher gains than similar field effecttransistor (FET) circuits and is not as noisy as metal-oxide semiconductorfield effect transistor (MOSFET) approaches. One of the principle disadvan-tages when considering neuroprosthetic applications, particularly microelec-trode recording, is the base current. This is very large in comparison toMOSFETs and causes polarization and electrode drift. Additionally, BJT cir-cuits typically consume more power than FET circuits. The BJT has generallybeen superseded by MOSFETs in digital logic circuitry, due to far lowerpower requirements of the latter.

13.1.5.3 The MOSFETAfield effect transistor consists of a channel between source and drain, controlledby a gate. In the depletion-type FET, a normally conducting channel is restrictedby applying a voltage to the gate, and in the enhancement-type FET a normallynonconducting channel is made to conduct by application of a voltage to thegate. In the metal-oxide semiconductor field effect transistor (MOSFET), the

Figure 13.2 (a) Diode in cross section. (b) Plan view (layout). Not to scale.

metal

oxide

p

np

(a)

(b)



gate electrode is separated from the channel by a thin insulating layer of silicondioxide (Figure 13.3). The properties of the device, for a particular fabricationprocess, are primarily determined by the width and length of the channel (theratio W:L being known as the aspect ratio). These are usually specified in

λ(lambda) units; one

λ unit being half the minimum feature size of the process.This is to facilitate scaling of the design as fabrication technologies improve.

Circuits can be constructed from n-channel MOSFETS (n-MOS), p-channelMOSFETS (p-MOS), or both types (complementary MOS, or CMOS). CMOSis implemented using only enhancement-type transistors. Broadly speaking,MOS approaches exhibit much lower power consumption that BJTs, but aregenerally slower (p-MOS is the slowest and CMOS the fastest of the three).That said, CMOS devices now effectively challenge BJTs in all but the mostdemanding of applications. In analog terms, MOS solutions suffer from flickernoise, which is inversely proportional to frequency and is significant in thefrequency range exhibited by neural signals (below 10 kHz); p-MOSFETs areslightly less noisy than n-MOSFETs. However, they also exhibit very highinput impedances and exceedingly low (gate) bias currents in comparison toBJTs. An additional characteristic of MOSFETs is the threshold voltage. This isthe voltage that must be applied between gate and source (VGS) in order forthe transistor to operate (“turn on,” in CMOS terms). IC fabrication processesthat mix CMOS and BJTs are now increasingly available. Known as bi-CMOS,this technology allows the designer to select the most appropriate solution totheir problem by mixing MOSFETs and BJTs as required.

13.1.5.4 The JFETThe junction field effect transistor (JFET) is available only as a depletion-type FET. It operates on the basis of a reverse-biased diode junction, whichcloses off the conducting channel as the bias is increased. The principleapplication of the JFET is in the input stages of amplifiers; they exhibit betterinput characteristics than BJTs and better noise performance than MOSFETs.

13.2 Elements of a systemDiverse designs are employed in neurotechnology. This section introducesthe reader to the circuit elements that have been employed and also to one

Figure 13.3 Structure of a p-channel enhancement-type MOSFET. The channel liesbeneath the gate electrode (G). Applying a negative voltage to the gate inducespositive carriers (holes) in the channel region, connecting the two p-doped regions.Not to scale.

S G D

p p

n

metal

oxide

semiconductor



or two likely candidates. The intention is to enable the identification ofcomponents that may be implemented in a system and the deciphering ofsystem block diagrams or circuit diagrams. The treatment focuses first onthe special case of the neuron–transistor junction, before discussing sequen-tially: CMOS circuits in general, general system building blocks (such asoscillators), circuits specific to recording (e.g., preamplifiers and filters), cir-cuits specific to stimulation (e.g., current output DACs), and concluding withbrief discussion of elements more specific to CNS, PNS, and retinal prosthe-ses. Due to overlap, the reader will find that some topics discussed in earliersections are also relevant to later sections. Najafi23 has provided a veryreadable introduction to neuroprosthetic systems incorporating MEMSdevices, which is illustrated by specific examples. See also Chapter 6.

13.2.1 The neuron–transistor junction

Simple FET follower circuits have been used as headstage buffer amplifiersfor conventional microelectrodes in the past; their purpose being to providean interface (impedance match) between the very high impedance recordingsite and relatively long cables with significant parasitic capacitance thatwould otherwise attenuate the signals of interest. One logical approach forintegrated microelectrode arrays would, therefore, be to integrate FET fol-lowers directly beneath metal recording sites, such that the gate conductorof the transistor formed the microelectrode itself. Such devices have beenreported,43 although design is complicated by a number of problems, includ-ing the need to provide noble metal gate metallization and ionic contami-nation of the gate.43 Additionally, follower circuits typically have a gain ofslightly less than unity so the implementation of a follower circuit as thefirst amplification stage introduces additional electrical noise while slightlyattenuating the signal of interest. Given these problems it may be preferableto position recording sites slightly away from the amplification circuitry,especially as modeling studies have suggested that, on the length scalestypically found in probes for CNS applications, parasitic capacitance andcross-talk can be kept at a manageable level even when multiple intercon-nection layers are implemented.16

Research led principally by Fromherz and Offenhäusser44–48 has, how-ever, focused on this type of interface. In this particular approach, there isno metallization on the gate of the FET used in recording. Stimulatingelectrodes are fabricated by thinning the insulation film above doped con-ducting tracks in the substrate. This approach is illustrated in Figure 13.4.The main advantage of this approach is that the electrode sites on the arrayare chemically virtually indistinguishable from the rest of the surface. Thishas been put to good use in terms of surface modification for patterningand studying cells in culture and is potentially interesting for biocompat-ibility purposes. Other advantages are that there are no problems withexposed fabrication layers at electrode sites (e.g., adhesion layers beneathnoble metal films or internal laminations of insulating/passivating films)



and no concerns about long-term stability of metal electrodes or electro-chemical activity at low frequencies, such as baseline drift, thus potentiallysimplifying fabrication. At present, this approach appears only to be usefulfor in vitro applications; however, studies into their electrical propertieshave been undertaken,48–50 and the advantages noted make a brief summaryof these relevant to the present topic.

Flicker noise in MOSFET circuits is a significant design considerationfor neural signal recording. When a metal recording site is used, especiallyat a distance from the MOSFET, then the designer can implement relativelylarge transistors to reduce flicker noise problems. Najafi,14 for instance, used145-µm-wide, 10.4-µm-long transistors; Ji,51 120-µm-wide, 15-µm-long tran-sistors. (In both cases, input p-MOS transistors were used as they are lessnoisy than n-type devices.) With a direct neuron–transistor junction, the gateregion (hence, the size of the transistor) is limited to the desired area of therecording site. Offenhäusser et al.46 reported gate dimensions of 28

× 12 µmdown to 10

× 4 µm. In this case, recording is limited to neural signals ofgreater than about 100 µV amplitude. In comparison, Bai et al.52 havereported overall noise levels as low as 20 µV for very complex three-dimen-sional recording arrays, and it is anticipated that noise levels can be reducedto below those of the metal recording sites themselves. In the in vitro situa-tion, however, neurons placed or grown directly over FETs can couple tightlyto the substrate. In this case, recorded signals can rise from several hundredmicrovolts to millivolt amplitudes. Such sealing to metal microelectrodesappears to be difficult to achieve. Unfortunately, when considering chronicneural implants, the microelectrode structure is typically surrounded byconnective tissue that will prevent this type of seal from being achieved.53–55

The electrical mechanism by which the nonmetallized stimulation sitesof these devices couple to cells placed or grown over them has been stud-ied.48–50,56 As may be expected, this is primarily a capacitive mechanism,complicated by the placement and coupling of the cell body to the substrate.This contrasts with metal microelectrode sites, however, in that when stim-ulation is applied using the latter, they have to be operated in a charge-balanced pseudo-capacitive regime in order to prevent degradation of the

Figure 13.4 Neuron–transistor junction and capacitive stimulation electrode.44–48 Notto scale.

FET

cell

stimulationsite



site through electrochemical activity. While the designer suffers from thisdisadvantage with metal microelectrode stimulating sites, this is compen-sated for by a number of advantages. The capacitance of these sites cangenerally be made to be greater than with insulated sites. The effectiveseparation of the plates is typically a few angstroms3 due to the electrochem-ical nature of the interface, surface roughening techniques can be used toincrease the effective plate area of the capacitor, and multivalent oxides (e.g.,of iridium or tantalum) can be used as the interface layer. All these consid-erations allow high charge densities to be achieved, resulting in more flexiblestimulation options.

13.2.2 General

A number of general considerations must be kept in mind when designingCMOS circuitry for implantable MEMS for neuroprosthetics. First, capacitorsimplemented on-chip tend to take up relatively large areas of silicon, whichis particularly valuable in PNS applications. Neural signals are of relativelylow frequency compared to those usually encountered in modern electronics,and relatively large capacitors are generally required to create appropriatefilter circuits. Furthermore, while it is often possible to fabricate adjacentcapacitors that match relatively closely on an IC, absolute tolerances tend tobe poor. Where precision is required, devices may be laser trimmed, but thisadds considerably to the cost of the process. As a consequence, capacitorsare best avoided when possible.

Similarly, resistors are difficult to fabricate. Some processes provide solu-tions to the creation of high value resistors on chip. Where this option is notavailable, the designer has to resort to the unsatisfactory solution of a verylong polysilicon track, and this is only an option for low value resistors. Toovercome this problem in the design of integrated circuits, the active loadis used. This is a transistor biased on by connecting the gate to the drain pin(Figure13.5c). This acts like a resistor, the exact performance being control-lable by the choice of aspect ratio.

Transistor-level circuit diagrams can be quite daunting at first sight;however, they generally consist of a few common motifs. These are repro-duced in Figure 13.5, using shorthand notation for enhancement-type n-MOSand p-MOS FETs. Note that source and drain are not distinguished; this isbecause the transistors will normally be symmetrical and the substrate biasis taken care of in very-large-scale integration (VLSI) design (for discretedevices, the substrate may be biased to source or drain).

Following from the active load configuration, the single transistorsource follower and the common source amplifier are shown in Figures13.5d and e. In the case of the source follower, the output will approximatethe input, the gain being close to but less than unity provided that the loadresistance is much greater than 1 divided by the transconductance of thetransistor (transconductance, gm, expressed as the ratio of output current tothe input voltage). In the common source amplifier, the current flowing



through the drain resistor is controlled by the gate voltage (id = gmVGS).Typically, the gain of such a configuration will be much lower than thatachievable using a common emitter BJT configuration. Note that in eachcase, an active load can be used, and closely matched transistor character-istics provide the designer with more options than would be availablewhen using discrete devices.

A further use of the availability of matched transistors is the currentmirror circuit (Figure 13.5f). Because the characteristics of these transistorsare almost identical and they are both biased by the same voltage, thenthe current flowing through transistor A must be the same as that flowingthrough B. Notice that by changing the aspect ratio of one of the transistorsit is possible to have a multiple of the current flowing through the other.

Figure 13.5 Common motifs in CMOS analog circuits. Shorthand notation for (a) n-MOSFET; (b) p-MOSFET; (c) active loads; (d) source follower with active load; (e)common source amplifier; (f) current mirror; (g) long-tailed pair (p-MOS) with cur-rent minor active load; and (h) transmission gate.

g

d

s

g

d

s

(a) (b)

+

-

+

-

(c)

+

0

Vin

Vout (d)

+

0

Vin

Vout (e)

(f)

+

0

Vout

Vin+ Vin-

(g)

x

x

signal(h)



This is often made use of in analog circuits; a single biased transistor isused to generate a reference current and a series of mirroring transistorsact as current sources for different parts of the circuit. Thus, variations insupply to one part of the circuit are mirrored throughout and correctoperation is maintained.

The most common input arrangement for an IC amplifier is the differ-ential pair configuration (Figure 13.5g); this is also known as a “long-tailedpair.” In this case, p-type MOSFETs have been used as the input transistors.The current source in the tail would typically be implemented as part of acurrent mirror, or as a biased transistor. A current-mirror load is shown inFigure 13.5g. The effect of this is that both sides of the amplifier contributeto the output, despite its being taken from only one arm. It is at this pointthat additional transistors may be added to adjust the gain of the first stageand introduce feedback within the amplifier design itself.

One final analog building block of interest is the CMOS transmissiongate (Figure 13.5h). In this case, the MOSFETs are acting as analog switches.They will either be high impedance (off) when x is the most negative voltagein the circuit or low impedance (on) when x is the most positive. The con-trolling signal would typically be a logic signal, so the analog signal beingswitched would have to lie within the voltage range used for logic. In thecase of an implantable system, one would endeavor to minimize the numberof power rails and run analog and digital parts of the circuit from the samesupply, although this is not without its drawbacks. Transmission gates canbe implemented in CMOS, n-MOS, or p-MOS technologies, although thereis no bipolar equivalent. If CMOS is not available, the properties of the gatewill be degraded by the threshold VGS required for the transistor to turn on.The transmission gate itself can form the basis of the analog multiplexer andmay also be used to discharge charge build up or apply test signals to theinputs of amplifiers. One must be aware, however, of capacitive feedthrough,both from input to output and from gate to channel.

Compared to analog circuitry, CMOS logic is much simpler to under-stand. CMOS gates are shown in Figure 13.6. More complex gates are createdby adding combinations of transistors in series and in parallel. The functionof the circuit can be determined by treating each transistor as a switch, withp-type MOSFETs being “on” when a logic 0 is applied to the gate, and n-type being on when a logic 1 is applied. Complex logic functions are thusbuilt up by implementing series and parallel combinations, for each seriescombination in the p arm, there must be an equivalent parallel combinationin the n arm, and vice-versa. Feedback from output to input would usuallyindicate a latch function of some type.

13.2.2.1 Clocks and oscillatorsDigital logic will normally require a clock signal to control it; this can be amultiphase clock, with several different signals being distributed across thecircuit to effect appropriate control, and it may also be necessary to ensurethat clock edges do not overlap so it can be guaranteed that certain transistors



turn off before others turn on. The simplest form of oscillator makes use ofthe inherent delay present in logic circuits. A string of inverters, with theoutput fed back to the input, will oscillate at a frequency related to the totaldelay through the chain. More common, however, is the relaxation oscillator,which comes in various forms based on the same principle. A capacitor isslowly charged up and then rapidly discharged, creating a sawtooth wave-form which is shaped to a square wave on the output. Figure 13.7 shows a

Figure 13.6 CMOS logic gates: (a) inverter, (b) NOR gate, (c) NAND gate.

Figure 13.7 Relaxation oscillator formed from two inverters.14,39



simple relaxation oscillator based on two CMOS inverters. Notice that whendesigning with discrete components, it is possible to use current sources andcurrent mirrors to control the current flowing into a capacitor and hence thefrequency of the oscillator. Once a square wave signal has been generated,it can be divided down to create the necessary clock phases. Inclusion ofinverter chains in this logic can ensure that edges do not overlap.

Given the caveats regarding implementation of capacitors in integratedcircuits, it is apparent that such circuits cannot be relied on to oscillate atprecisely the desired frequency nor to be as stable as those that incorporatequartz or similar resonators. Thus, these circuits cannot be relied on forprecise timing. One use for such oscillators has been the control of analogmultiplexers for multichannel recording probes.13,51,58,59 Here the clock signalof the oscillator has been regenerated off-chip in order to decode and demul-tiplex the recorded signals. Thus, frequency differences between individualoscillators could be accounted for if necessary. Peeters et al.59 implementeda two-wire system for recording probes, which obviously required an on-chip oscillator for multiplexing.

Recent designs have, however, tended to rely on off chip clock sig-nals,60–62 and Ji’s design51 implemented a serial communications protocol thatenabled the clock to be regenerated from the data received. This wasachieved by using multilevel logic signals (5 V on the I/O line toggled thedirection; 4 V was a logic 1, and 2V a logic 0; and bits were separated by a0-V period, so that the logic received a series of pulses that could be usedto regenerate the clock). Injectable stimulators31,32 can use a similar approach,whereby the RF carrier (of a few megahertz) used to power and communicatewith the device provides it with its primary clock signal.

The main problem with clocked logic, notably in recording applications,is that of clock coupling through to the analog section. For example, Baiet al.52 reported about 40 µV of noise on recovered demultiplexed signals,compared to 20 µV of noise with nonmultiplexed signals. Coupling can bethrough stray capacitance or through power line glitches; it is difficult toseparate digital and analog sections of circuitry on ICs in the manner thatis common for discrete circuit design, and decoupling capacitors are notavailable. Various strategies are available for ameliorating this problem, themost basic being good circuit design, care when laying out the circuit, andgood use of extraction and simulation from the layout.64

13.2.2.2 Power-on reset and mode controlIt is recommended that logic circuits be supplied with a reset signal to ensurethat they can be put into a known state when desired. This is normallyperformed by supplying an additional reset line to the circuit, but whenconcerned about the number of wires attached to an implant it would bedesirable to eliminate a dedicated reset connection if possible. The first wayto do this is to ensure that a reset signal is supplied on power-up. This isnormally performed using a capacitor and flip-flop circuit. A transistor orlow resistance path to ground is placed so as to ensure that the flip-flop



powers up in one particular state (reset signal active). When power is sup-plied to the circuit, the capacitor will begin to charge, thus providing adelayed signal to change the state of the flip-flop, inactivating the resetsignal.14,51 A few caveats regarding this particular approach should be noted.The first is that the power supply must ramp up to maximum much morerapidly than the capacitor charges. This could be a problem with implantedintegrated circuits where not only is the size of the capacitor limited byavailable silicon real estate, but the current available from an implantedpower source may also be limited.

Where an external clock signal has been made available, designs haveincorporated this into the power-on reset circuitry such that it is the firstpulse on the clock line that inactivates the reset signal.60 Designs for RF-powered injectable stimulators and retina prostheses31,35,36,65 generally appearto incorporate a synchronization sequence within the data transmission pro-tocol that effectively resets the device. This approach can be effective whenfunction is relatively limited; however, state machine design can becomeburdensome and demanding on silicon real estate. Additionally, withoutcareful implementation of power-up reset circuitry, one risks the possibilitythat the device will power up in an unsafe state, and the safe power cyclingof stimulation circuits must be considered (see, for example, Loeb et al.31).

Because CMOS circuits, unlike bipolar circuits, can operate over a rangeof supply voltages, this provides the designer with further options in reset-ting or changing the operating mode of the circuit. Self-test features havebeen implemented13,58 that are activated by raising the supply voltage abovenormal operating voltage. Once again, this can be implemented by a flip-flop activated by a transistor attached to the supply rail in such a way thatit does not reach threshold and turn on until the supply rail is raised to aspecific voltage. Ji and Wise58 also introduced a multilevel logic systemwhereby a comparator was used to detect a 5-V signal on the signal I/Oline. This enabled the design to toggle between send and receive modeswithout requiring an additional signal line.

13.2.2.3 IntegratorsThe integrator is a useful circuit element that functions as the name suggestsby integrating an input voltage signal. Electronic integrators are not perfect,being limited by finite supply rail voltages, currents, and leakage; they are,in fact, a form of low-pass filter. The designer has two basic forms to choosefrom, both of which use a capacitor as the integrating element.40

A transconductance integrator involves conversion of the input voltageto a current, usually through transistor action, which is used to charge acapacitor. The output voltage is measured across the capacitor. A disadvan-tage of this circuit is the need to place additional circuit elements in theoutput circuit of the converter to provide feedback control, which can affectnoise performance or dynamic range. For this reason, operational amplifierintegrators are often preferred for such applications; this is essentially anoperational amplifier circuit with a capacitor in the feedback path.



Beyond filtering, the principle application for integrators is to providesawtooth signals and reference slopes (as elements within relaxation oscil-lators, for instance). These may be incorporated into analog-to-digital con-verters or voltage-controlled oscillators for use in phase-locked loops.

13.2.2.4 Phase-locked loopsThe phase-locked loop (PLL) is used to synchronize a system clock signalwith an incoming clock signal for which only the approximate frequencyrange is known. They can also be used as frequency multipliers to synthe-size high-frequency clocks from low-frequency reference signals. The prin-ciple use of PLLs in neurotechnology has been in decoding the complexmultiplexed data signals generated by recording devices.66 The PLL (Figure13.8) consists of a voltage-controlled oscillator (VCO), the control beingprovided through an integrator element which ramps up at a rate deter-mined by the input voltage, before being reset. The phase of the shaped(square wave) output of the VCO is compared with input signal, and thedifference is fed back as a voltage through a low-pass filter to the control-ling input of the VCO. The output of the PLL locks onto the reference signalafter a period of time determined by the filter in the feedback path; there-after, synchronization can be maintained even if some transitions of thereference signal are missed.

13.2.3 Recording

Further to the general circuit elements discussed in the proceeding section,a number of elements have more specific application in the recording ofneural signals through MEMS-based microelectrode devices. These are, inparticular, preamplifiers, filters, switched-capacitor filters, and analog-to-digital converters.

13.2.3.1 Signal preamplifiersOne of the earliest goals of MEMS-based electrodes for neurophysiologicalapplications was to be able to integrate signal preamplifiers as close to therecording site as possible, thus reducing as far as possible the problems ofexternally induced noise.4,43 Beyond the general requirements of good lownoise design, good power-supply rejection ratio (PSRR, a measure of theimmunity of the amplifier from noise on the power-supply lines), and lowpower consumption, the designer is faced with additional problems not

Figure 13.8 Components of a phase-locked loop (PLL).

signal Phasecomparator VCO

low-pass filter

output



normally encountered in amplifier design — specifically, baseline (DC) driftof microelectrode recording sites, polarization of the recording site, andcharge build-up on the MOSFET gate.

It is known that even the passage of very small (fA) DC currents acrossmicroelectrode sites can result in offset signals that are orders of magnitudelarger than the signals of interest.14 This generally precludes the use of low-noise BJTs or JFETs as the input stage of preamplifiers, because the offsetvoltages induced would tend to saturate the input stage of the amplifier,and current flow may have a deleterious effect on the electrode site andsurrounding tissue. The FET solution then suffers from the possibility ofcharge build up on the gate of the transistor. An additional problem, undercertain circumstances, is optically induced currents.67 The designer mustcompromise the need for a high-input impedance, which must be at leastone order of magnitude greater than that of the site, against the requirementsof low gain at DC and a path of known impedance for bias currents anddischarge of charge build up. In discrete circuits this can be solved using ahigh-value resistor and a relatively high supply voltage (e.g., ±12 V) to avoidsaturation of the first amplifier stage; more severe filtering would be per-formed by subsequent circuits.68

A biased diode can be employed to provide the appropriate input resis-tance required,13,69 often in conjunction with a transmission gate that can beturned on to provide a discharge path to ground, if necessary. At least oneearly design implemented a capacitor to decouple the recording site.67 This,however, required considerable silicon real estate for the 30 pF capacitorsthat were required, compared to capacitors of 3 pF or less employed in DC-coupled designs with filters.13,58 More recent work by Bai and Wise69 hasreviewed a number of possibilities for DC baseline stabilization for a closed-loop amplifier without any filtering. In this case, a closed-loop amplifier wasdesired, as it was seen as providing more stable gain characteristics.

Furthermore, it must be recognized that the high-frequency cut-off isequally of importance when considering the noise performance of the sys-tem. Normally, however, this is less problematic than the low-frequency cut-off, as it can be implemented with a relatively small loading capacitor in theoutput stage (see, for example, Ji and Wise58) or through design (for example,Takahashi and Matsuo67 used the gate capacitance of the input transistor toachieve a high-frequency cut-off). Further studies on noise performance forintegrated neural signal preamplifiers have recently been published.69–71

13.2.3.2 FilteringAs previously stated, filter design is compromised by the problems of con-serving silicon real estate and fabricating low-tolerance capacitors on-chip;in such cases, the designer may find it desirable to avoid continuous timeactive filters where possible. Takahashi and Matsuo,67 however, used largedecoupling capacitors between recording site and amplifier, whereas Bai andWise69 have attempted to obviate the need to attenuate low-frequency com-ponents in the preamplifier by careful input stage design. Otherwise, the



preferred approach in preamplifier design has been to implement a diode-capacitor filter58,60,72 (Figure 13.9); the equivalent resistance of the diodes isrelatively high, enabling the designer to use correspondingly small capaci-tors to achieve the desired low-frequency response. This filter design wouldnormally provide negative feedback from the output stage of the amplifierto the input stage.

The principle alternative to continuous time filtering is the switchedcapacitor filter.39 This is implemented with CMOS switches, as indicated inFigure 13.10. The switches are operated at a frequency much greater thanthe highest frequency in the signal, and the circuit acts as an integrator (Vout

= f0c1/c2 ∫ Vindt). It can therefore be used to substitute continuous time inte-grators in filter circuits. Because the transfer function of the circuit dependson the ratio of the two capacitors, rather than their absolute values, it lendsitself to IC fabrication where the relative values of two components can becontrolled to a much greater degree than their absolute values. It is alsopossible to tune the filters thus realized through the clock frequency.

Switched capacitor filters do, however, have a number of disadvantagesfor the present application. First, there is the problem of clock feedthrough,where transients appear in the output signal at the frequency of the clocksignal used. Furthermore, any noise in the input signal at or near the clockfrequency will be aliased down and appear in the output signal within thepass-band. Finally, in addition to a generally reduced dynamic range com-pared to continuous time-active filters, the switched-capacitor filter willtypically consume more power. This is due to the fact that power dissipationin CMOS logic circuits is strongly related to clock frequency. For these rea-

Figure 13.9 Diode-capacitor lowpass filter that has been implemented in severalintegrated neural signal preamplifier circuits.58,60,72

Figure 13.10 Principle of the switched capacitance filter.

+

0

Vin Vout

+-

0

Vin

Vout

C1

C2x f0 x



sons, the switched capacitor filter is probably not appropriate in the earlystages of a recording system; however, it could potentially be used to goodeffect in some applications.

13.2.3.3 Analog-to-digital convertersThe principle problem with multiplexing analog signals onto a single wirefor transmission to a centralized control unit of a neuroprosthesis is that thesignals, although amplified, are still susceptible to interference. For thisreason, conversion of the signals to digital format is desirable, and thedesigner has a number of options by which to implement this.

At this point it will be assumed that a full reproduction of the recordedsingle unit action potential is required. Normally one would use an upperfrequency of interest of 10 kHz, but for a best case when considering con-verter design an upper frequency of 3 kHz will be assumed here.73,74 In thiscase, the converter would have to sample the signal at least every 0.1 msec(10-kHz sampling frequency). For 100 recording sites, this would require a1-MHz converter, or several converters operating at lower frequencies. Theproblem is not, therefore, as simple as it first appears to be. A variety ofanalog-to-digital converter (ADC) designs can be selected for this purpose,the most common and promising being the successive approximation ADC,the flash converter, the sigma–delta converter, and integrating converters.The function of these being outlined below. These converters will generallyneed to be combined with sample and hold circuits to stabilize the inputsignal during the conversion process. These contain capacitors that arecharged from the input signal through a digitally controlled switch.

13.2.3.3.1 Successive approximation ADC. The operation of the succes-sive approximation ADC is illustrated diagrammatically in Figure 13.11. The

Figure 13.11 Illustration of the operation of a four-bit ADC. The input signal is shownas a dashed line, and the test value of the DAC is shown as a heavy line. In the firstcycle (a), the most significant bit (msb) of the DAC is set and the output comparedto the input signal. The input signal is greater; therefore, the bit remains set at theend of the step. Then (b) the next bit of the DAC is set. The input signal is less thanthe output of the DAC so this bit is reset. This is repeated in steps (c) and (d) to givea final value of 1010 after four steps.

1000

1000

1010

1010

1000

1100

1010

1011

(a)

(b)

(c)

(d)

input signal

DAC final value

DAC test value

msb

msbtime

V



converter is comprised of a comparator and a digital-to-analog converter(DAC), the output of which is compared to the analog input signal. The firstbit of the DAC is set and the output compared to the input signal. If theinput is less than the output of the DAC then the most significant bit is reset.This sequence is then repeated for the next most significant bit and theremaining bits in sequence. The number represented on the inputs to theDAC is then the output of the ADC. This is a clocked circuit, and powerdissipation will be affected by clock frequency; this in turn will depend onthe number of bits used in the converter and the desired conversion time.

13.2.3.3 Flash ADC. The conversion time of the flash ADC is limitedonly by signal delays through the circuit and the frequency response of theanalog components. It consists of a reference resistor chain and a number ofcomparators related to the number of quantization levels desired. If an eight-bit result is required, then 255 separate reference voltages are generated and255 comparators are used to compare the input signal to each of these. Theoutput of the comparators is then converted to an eight-bit number usingconventional combinatorial digital logic. The power dissipation of the con-verter and area of the circuit are related to the number of bits required inthe output. The converter will also be relatively inaccurate due to processvariations. There are alternatives to full-flash converters that incorporatepartial flash conversion of the signal. These designs trade reduced compo-nent count for speed.

13.2.3.4 Sigma–delta ADC. The sigma–delta converter has recentlygained popularity due to the increasing speed of CMOS ICs. There arevarious designs, but they essentially consist of a single comparator and anintegrator. In the simplest form, a number of pulses are supplied to theinput of the integrator over a specific time period. When the output of theintegrator reaches the value of the input signal then the number of pulsessupplied provides the output of the converter. This is the opposite of theflash converter. For an eight-bit converter, for instance, 255 clock pulsesare required to effect the conversion. Sigma–delta converters are generallythe most accurate converters available; however, the conversion timesavailable and high clock frequencies suggest that they are not suitable forthe present application.

13.2.3.5 Integrating ADCs. Integrating ADCs are, perhaps, the mostuseful form of ADC for the present application. The simplest form of inte-grating ADC consists of a comparator, counter, current source, and capacitor.The counter is started at the same time that the current source is applied tothe discharged capacitor. When the voltage across the capacitor is equal tothe input signal, the counter is stopped and the number it contains representsthe input signal level. The advantage of this approach is that the two halvesof the converter can be separated. The capacitor and comparator can beimplemented near the recording site, and the counter can be implemented



on the central controller, which may not have so many design restrictionsas remote components of the system. Signals from several different sites canbe multiplexed together as pulse-width modulated (PWM) signals for trans-mission via a single connecting wire. The speed of the converter is essentiallylimited by the clock frequency, although the analog circuitry will itself playa part at high clock frequencies. Process variations can also limit the accuracyof the converter. Use of dual slope converters can help overcome this prob-lem. In this case, the capacitor is charged for a fixed period of time from acurrent source that is proportional to the input voltage. The capacitor is thendischarged using a fixed current source; the discharge time will be propor-tional to the input voltage.

13.2.4 Stimulation

Following this review, only one major circuit element remains that is specificto MEMS-based stimulators, and that is the stimulus delivery circuit andassociated digital-to-analog converter. The need for a stable power supplymust be kept in mind, however.

13.2.4.1 Voltage-output DACsThe output stage of a stimulator is normally designed to deliver biphasiccharge balanced pulses. For this purpose, current output DACs, as describedin the following section, are most appropriate. Applications for voltage-output DACs would include ADCs, as described in the previous section, orthe control of current sources. Note that current output DACs can be usedin these applications with an appropriate current to voltage output stage.Two common voltage-output DACs are shown in Figure 13.12, one basedon a summing amplifier and one on the R-2R ladder network.

13.2.4.2 Current output DACsThe conventional form of the current output DAC is shown in Figure 13.13.It consists of a series of scaled current mirrors, with MOS switches to connectthem to the electrode site. In Figure 13.13 both the source and sink halvesof an eight-bit converter are shown, although not all the bits are shown. Thisformat has been implemented for a number of different designs.37,60,61 Nor-mally, only one or a few converters would be implemented, with the outputsmultiplexed onto different electrode sites.

An interesting bi-CMOS version has been implemented by Kim andWise.75 In this version, scaled transistors supply current to the base connec-tion of a BJT. The current flowing through the BJT is proportional to the basecurrent, thus a converter was realized with a much reduced transistor count,the current mirroring transistors being rendered unnecessary. In addition,the circuit was designed for reduced power operation. Tanghe and Wise61

also published an interesting circuit, where the same current source pathwas used for both source and sink current mirror circuits. This should ensurecloser matching between source and sink currents.



Another option to ensure balanced charge operation has been to use asingle-current source DAC and switch this between active and referenceelectrodes using a H-bridge style circuit.36,76 Suaning and Lovell36 also imple-

Figure 13.12 Four-bit DACs: (a) based on a summing amplifier, and (b) based on theR-2R ladder network. Both examples show inverting amplifiers so the outputs willbe negative. D3 is the msb; D0 the lsb.

Figure 13.13 Eight-bit current output DAC with source and sink outputs. Transistoraspect ratios are for illustration only. Note that the data bits to the source transistors(DPx) are inverted. Not all of the transistors have been shown.

-+

Rf

R

2R

4R

8R

D3

D2

D1

D0

0

Vout

(a)

-+

0

R

Vout

0

VrefR R R

2R 2R 2R 2R

2R

(b)

D0D1D2D3

-

0

source

sink

11

21

1281

641

11

21

1281

641

DP0DP1DP6DP7

DN0DN1DN6DN7



ment decoupling capacitors on the output of the bridge circuit to isolate theelectrode sites from DC bias.

An alternative approach has been described for injectable stimulators.31,32

In the system described by Loeb et al.,31 a steady-state polarization is main-tained between capacitive working and reference electrodes. Stimulation isby way of a current pulse (1.5 or 15 mA), which discharges the capacitor. Asmaller current (0.01 or 0.1 mA) then recharges the system. Stimulation isprincipally effected by the high-amplitude pulse, and charge balance is main-tained due to the longer recharge period at lower current.

13.2.5 Elements specific to CNS prostheses

The design constraints on prostheses located within the skull are a little morerelaxed than for prostheses that have components distributed throughoutthe PNS. For this reason power regulation and communication are treatedas elements specific to PNS prostheses, because greater problems are antic-ipated for PNS than CNS prostheses in this respect. As to elements specificto CNS prostheses, the more complex and resource-hungry circuits discussedare more likely to be implemented in CNS prostheses than in PNS prostheses.At present, technologies are not sufficiently advanced to be able to pick outparticular circuit elements that are specifically applicable to CNS prostheses;those discussed tend to represent elements that are desirable for both CNSand PNS applications. More specific elements will appear as specializedsignal processing elements, as MEMS-based neuroprosthetic systems becomemore application orientated.

13.2.6 Elements specific to PNS prostheses

Unlike components located within the skull, it is desirable that MEMS-basedcomponents located in the PNS should have as few leads attached as possi-ble. This leads to one aspect more peculiar to PNS prostheses and retinaprostheses than to CNS prostheses — specifically, radiofrequency (RF) com-munication and power transmission. This has been a particular factor ininjectable stimulators31,32 and retinal prostheses. The unique problems asso-ciated with this approach stem from its combination with MEMS and theresulting restrictions on size and location, particularly with injectable stim-ulators.36,65,77,78 Within this section, however, the aspects addressed will bethose of circuit design, particularly power regulation and approaches tocommunication.

13.2.6.1 Voltage regulationVoltage regulation of some degree may be required for implants withwire connections, particularly if these incorporate sensitive circuitry andsome signaling is performed by voltage changes on the power supplylines.13,58 RF-coupled power31,36,37,63 will be rectified and a large smoothingcapacitor will used across the input of the regulator. Voltage regulators



operate from a reference voltage that will usually be in the form of azener diode. Ziaie et al.32 used zener diodes with BJTs, where the forwardvoltage drop of the base-emitter junction is known (0.7 V) to form a basicvoltage regulator. More complex voltage regulators incorporate an ampli-fier of some sort with a high current output stage. The basic principle isillustrated in Figure 13.14. This amplifier could be as simple as a long-tailed pair, but note that the input supply has to be smoothed sufficientlyby the capacitor to ensure operation.

Voltage regulators are going to be temperature hot spots on any die,because they will incorporate one or more components that have a largevoltage drop across them with a large current flowing. This means thatdesign and location of power regulation circuits needs to be carefully con-sidered. Not all parts of the circuit may require voltage regulation; wherestimulation is current controlled, for example, the rectified and smoothedsignal could be used (see, for example, Loeb et al.31).

Recently, an alternative to powering via RF coupling or implanted bat-teries has been proposed: transcutaneous optical powering.79–81 Optical pow-ering and even data may be supplied by an infrared (IR) LED array or laserand converted to electrical power through PIN photodiode arrays. Edell79

also presented a scheme by which recorded signals would be transmittedoptically from the implant. Obviously this scheme has also received partic-ular attention in retinal prosthetic applications.33,34

13.2.6.2 Communication schemesAlthough IR powering and data transmission has been suggested for wire-less implants,33–35,79–81 RF approaches appear to be more developed, prob-ably due to the history of this technique in prosthetic applications. Afavored approach to data transmission has been the use of an amplitude-modulated (AM) carrier.31,35,65 Fast switching of this carrier is achieved byfrequency modulating the original data and delivering this to a tuned(resonant) antenna circuit. When the frequency of the signal deviates fromthe resonant frequency of the circuit, a reduction in amplitude of thetransmitted signal results.

Figure 13.14 Principle of a voltage regulator based on an amplifier and referencesource. Note that the output will be a multiple of the reference (Vref) and that theinput voltage must be smoothed sufficiently to ensure that the reference is stable andthe amplifier can drive to at least Vout + 0.7 V.

+-

Rf

0

Vout

R1

Vin

Vref



Demodulation of the transmitted signal can be achieved by filtering toremove the carrier frequency, and the subsequent use of a comparator orSchmitt trigger.32,35 This is essentially a relatively low open-loop gain ampli-fier with some positive feedback. The positive feedback induces somehysteresis in the circuit, making it noise tolerant to some degree. Theinjectable devices described by Loeb et al.31 demodulate the transmitteddata by comparing short-term fluctuations in the received signal with alonger term average.

Some groups have used pulse-width modulation (PWM) techniques, orvariations thereon, to encode the data.37,65 In this situation, the ratio of thehigh (mark) period to the low (space) period of the AM signal provides theindication of a logic 1 or 0. The usual way to decode these signals is toimplement a clock, possibly derived from the RF carrier frequency,31 whichdrives a counter circuit. This would be used to count the time of a mark (orspace) in the incoming signal, and hence determine the value of the corre-sponding bit.

Clements et al.37,76 implemented a delay locked loop (DLL) and a varia-tion on PWM. Three taps from a voltage-controlled delay line were used todetermine whether or not the incoming signal represented a 1 or a 0. Theadvantage of this approach was that data transmission would have relativelylittle effect on the power level received by the implant, and a clock signalcould be derived from the incoming data. The DLL consisted of a string ofinverters, whereby the current controlling them (and hence the delay throughthem) was independently controlled. Thus, the total delay could be adjustedto the frequency of the incoming data in a manner reminiscent to the oper-ation of a PLL.

Manchester encoding has also been implemented.31,35,82 Logic levels arerepresented by a transition halfway through the bit; a logic 0 is representedby a low to high transition, and a logic 1 by a high to low transition. Clocksignals may be recovered from this encoding scheme, and it has the addi-tional advantage that the AM signal is low for exactly half a bit period, andhigh for the other half.

The data transmission protocol for such implanted devices has generallybeen of a broadly similar format. An initial reset or synchronization signalwould be followed by a binary address — the number of the device forwhich the subsequent data were intended, and a set of stimulation param-eters encoded in binary. The stimulation period either would be part of thesebinary-encoded parameters or would be signaled in some manner by thetransmitter (e.g., the time the signal remains high following the last word ofdata). Parameters would typically be programmed for every single pulse, aspreprogrammed parameters could become corrupted. If incomplete datawere to be received, this would be discarded until a new synchronizationor reset signal were received and the process would begin again.

Suaning and Lovell36 reviewed RF communications with neuroprosthesesin relation to their application, a retinal prosthesis. They implemented anapproach whereby parameters were set by counting down the number of



pulses transmitted in a particular packet. Different timing mechanisms werealso implemented to ensure that data were correctly received. Although elec-trode selection was implemented, device addressing was not. Suaning andLovell also implemented a reverse telemetry system that provided an indica-tion of power levels supplied to the IC and to the stimulation circuitry. Con-necting the supply voltage to the receiver antenna would cause a pulse thatcould be detected by the transmitter during quiet periods in the transmission.

Following completion of stimulation, the voltages of interest would beconnected to resistor–capacitor circuits. These would discharge, and, uponreaching a reference level, the telemetry pulse would be effected. Thus, thetime period between completion of stimulation and reception of the reversetelemetry pulse would indicate the levels of these voltages. These couldbe used as diagnostics, not only to confirm operation of the implant butalso as an indication the electrical characteristics of the tissue adjacent tothe electrodes.

13.2.7 Elements specific to retina prostheses

Most of the elements required by a retina prosthesis have been discussed inthe preceding sections. One unique approach to the retina prosthesis hasbeen to utilize arrays of photodiodes coupled to electrodes in such a waythat illumination of the photodiode would stimulate the adjacent tissue,33,34,83

power for the stimulation being derived from the incident illumination.Although the practical application of this approach has been demonstrated,at least one group feels that the intensity of illumination required to effectstimulation would be too great for long-term application and has recom-mended an array combined with an external power source for stimulation.34

This would have the added advantage of enabling the use of a higherresolution array.

13.2.8 Other elements

The use of flexible polyimide microelectrode devices10,11,84,85 prevents thedirect integration of electronic devices onto the microelectrodes themselves(a monolithic solution). Hybrid solutions, where different elements of asystem are produced using different technologies and then assembled, haveadvantages in terms of cost and speed of solution. Schuettler et al.12 have,for example, reported a hybrid system incorporating a micromachined poly-imide electrode array and interconnection system, three bare dice of com-mercially available ICs, and five passive components. A similar hybridapproach, although with custom-designed ICs, has been used in the devel-opment of a high-density connector system,86 and a flip-chip approach hasbeen proposed for contacting pincushion arrays.57,87 Another example of thehybrid approach is an RF transmitter developed for biotelemetry.63 Thisincorporated a surface-mount RF transistor with MEMS components (induc-tor, metal–polysilicon capacitors, and a polysilicon resistor).



13.3 Future directionsIt is clear that we need to know a lot more about the nervous system andlikely tissue reactions to MEMS-based devices in order for the technologyto achieve its full potential. One can expect that application of the currentgeneration of technology can show the way in this respect (e.g., see Refer-ences 9, 21, and 88). In addition, injectable stimulators and retinal prosthesisprojects are showing considerable promise in providing direct benefit topatients from current technologies.

One of the clear implications for the future is that the low-power circuitdesign techniques that have resulted from the demand for portable equip-ment will be taken up and implemented in neurotechnology. It also seemslikely that the widening access to bi-CMOS processes on a multiproject basis(e.g., through MOSIS in the United States and Europractice in Europe) willresult in increasing numbers of bi-CMOS designs, which are able to incor-porate of the advantages of both bipolar and CMOS technologies.

What is more difficult to predict is the form that future circuit designswill take: full custom, semicustom, monolithic with MEMS, hybrid, etc. Themechanical properties of silicon and the biocompatibility results achievedwith polyimide substrate microelectrode structures by the Fraunhöfer Insti-tute for Biomedical Engineering10,11 has led to further development of poly-imide-based devices, including penetrating arrays.85 In cases where the elec-trode substrate is not silicon, the fabrication process for the IC is notconstrained by enforced compatibility with the MEMS process.

The design problem appears challenging enough that full-customdesigns will, for the short term at least, provide the best solutions. Thedesigner is, however, increasingly provided for in terms of tools, technol-ogies, and intellectual properties that can be brought to bear on the prob-lem. The current industrial trend for System-on-Chip (SoC) designs thatincorporate analog and digital components on a single die should spin-offincreasing knowledge and improved technologies that can be taken up inneuroprosthetics.

Another area of investigation is powering and control of implants. RFhas tended to be dominant in the past, with IR a relative newcomer. Newand innovative approaches to this problem would undoubtedly be welcome.

As mentioned, increased understanding of the operation of the nervoussystem is essential. This is particularly important when the potential data-gathering capacity of MEMS-based multimicroelectrode devices is considered(e.g., one device with 100 recording sites, each sampling extracellular spikeswith information content to 10 kHz and eight-bit precision, implies a minimaldata rate of 16 Mbps and the necessary processing power to make sense ofthe incoming data). Obviously, data compression and interpretation have tooccur near the recording site; the best way to approach this problem will beinformed by experimental results arising from the existing technology.

In 1991, Peeters et al.59 presented a design that illustrates some of theingenuity that has gone into solving the problems presented. A probe with



ten recording sites was interfaced using only two wires: power and ground.The signal arising at each recording site was amplified and these were thenmultiplexed onto the input of a second amplification stage. An on-chiposcillator was used to drive the multiplexer. The output of this stage wascompared with a reference voltage; any signals above a preset thresholdindicated neural activity at the site. This was signaled through a currentsource that was connected across the power lines. The signal was pulse-width modulated to indicate the active channel. Thus, activity on channel 1would cause a current pulse of 10-µsec duration to be drawn from the powersupply, and activity on channel 10 would cause a 100-µs pulse; these pulseswere detectable by the external circuitry. The disadvantages of such a systemare quite apparent; however, the description provides considerable inspira-tion when dealing with the problems posed to neurotechnology.

AcknowledgmentsThe author would like to thank K. Bustamante, M. Hughes, D. Ewins, andthe Biomedical Engineering Group at the University of Surrey for access tofacilities and publications.

References1. Stamford, J.A., Ed., Monitoring Neuronal Activity: A Practical Approach, IRL

Press at Oxford University Press, Oxford, 1992.2. Llinás, R., Nicholson, C., and Johnson, K., Implantable monolithic wafer

recording electrodes for neurophysiology, in Brain Unit Activity During Be-haviour, Philips, M.I., Ed., Charles C Thomas, Springfield, IL, 1973, chap. VII.

3. Wise, K.D., Angell, K.B., and Starr, A., An integrated circuit approach toextracellular microelectrodes, IEEE Trans. Biomed. Eng., BME-17(3), 238, 1970.

4. Wise, K.D. and Angell, J.B., A low-capacitance multielectrode probe for use inextracellular neurophysiology, IEEE Trans. Biomed. Eng., BME-22(3), 212, 1975.

5. Normann, R.A., Maynard, E.M., Guillory, K.S., and Warren, D.J., Corticalimplants for the blind, IEEE Spectrum, May, 54, 1996.

6. Loeb, G.E., Walmsley, B., and Duysens, J., Obtaining proprioceptive informa-tion from natural limbs: implantable sensors vs. somatosensory neuron re-cordings, in Physical Sensors for Biomedical Applications, CRC Press, Boca Raton,FL, 1980, chap. 10.

7. Prochazka, A., Westerman, R.A., and Ziccone, S.P., Discharges of single hind-limb afferents in the freely moving cat, J. Neurophysiol., 39, 1090, 1976.

8. Prohaska, O.J., Olcaytug, F., Pfundner, P., and Dragaun, H., Thin-film multipleelectrode probes: possibilities and limitations, IEEE Trans. Biomed. Eng., BME-33(3), 223, 1986.

9. Meister, M., Pine, J., and Baylor, D.A., Multi-neuronal signals from the retina:acquisition and analysis, J. Neurosci. Meth., 51, 95, 1994.

10. Navarro, X., Calved, S., Rodríguez, F.J., Stieglitz, T., Blau, C., Butí, M., Valder-rama, E., and Meyer, J-U., Stimulation and recording from regenerated pe-ripheral nerves through polyimide sieve electrodes, J. Periph. Nervous Syst.,3(2), 91, 1998.



11. Stieglitz, T., Beutel, H., Schuettler, M., and Meyer, J.-U., Micromachined, poly-imide based devices for flexible neural interfaces, Biomed. Microdevices, 2(4),283, 2000.

12. Schuettler, M., Kock, K.P., Stieglitz, T., Scholz, O., Haberer, W., Keller, R., andMeyer, J.-U., Multichannel neural cuff electrodes with integrated multiplexercircuit, in Proc. 1st Ann. Int. IEEE-EMBS Special Topic Conf. on Microtechnologiesin Medicine and Biology, Lyon, October 12–14, 2000, p. 624.

13. Najafi, K. and Wise, K.D., An implantable multielectrode array with on-chipsignal processing, IEEE J. Solid-State Circuits, SC-21(6), 1035, 1986.

14. Najafi, K., Multielectrode Intracortical Recording Arrays with On-Chip SignalProcessing, Ph.D. thesis, University of Michigan, Ann Arbor, 1986.

15. Edell, D.J., Clark, L.D., and McNeil, V.M., Optimization of electrode structurefor chronic transduction of electrical neural signals, in Proc. 8th Ann. Int. Conf.IEEE Eng. Med. Biol. Soc., Houston, TX, 1986, p. 1626.

16. Najafi, K., Ji, J., and Wise, K.D., Scaling limitations of silicon multichannelmicroprobes, IEEE Trans. Biomed. Eng., 37(5), 474, 1990.

17. Normann, R.A., Visual neuroprosthetics: functional vision for the blind, IEEEEng. Med. Biol. Mag., January/February, 77, 1995.

18. Dobelle, W.H., Artificial vision for the blind by connecting a television camerato the visual cortex, ASAIO J., 46, 3, 2000.

19. Anderson, D.J., Najafi, K., Tanghe, S.J., Evans, D.A.. Levy, K.L., Hetke, J.F., Xue,X., Zappa, J.J., and Wise, K.D., Batch-fabricated thin-film electrodes for stimu-lation of the central auditory system, IEEE Trans. Biomed. Eng., 36(7), 870, 1989.

20. Rauschecker, J.P. and Shannon, R.V., Sending sound to the brain, Science, 295,1025, 2002.

21. Serruya, M.D., Hatsopoulos, N.G., Paninski, L., Fellows, M.R., and Donoghue,J.P., Instant neural control of a movement signal, Nature, 416, 141, 2002.

22. Kennedy, P.R., Bakay, R.A., Moore, M.M., Adams, K., and Goldwaithe, J.,Direct control of a computer from the human central nervous system, IEEETrans. Rehabil. Eng., 8(2), 198, 2002.

23. Najafi, K., Micromachined systems for neurophysiological applications, inHandbook of Microlithography, Micromachining, and Microfabrication, Vol. 2. Mi-cromachining and Microfabrication, Rai-Choudhury, P., Ed., SPIE Optical Engi-neering Press, Washington, D.C., and IEE, London, 1997, chap.9.

24. Bogart, T.F., Electronic Devices and Circuits, Merrill Publishing, Columbus, OH,1986.

25. Andrews, B.J. and Rushton, D.N., Lower limb functional neurostimulation,Baillière’s Clin. Neurol. Int. Practice Res.: Neuroprostheses, 4(1), 35, 1995.

26. Troyk, P.R. and Donaldson, N. de N., Implantable FES stimulation systems:what is needed?, in Proc. IFESS 2001, the 6th Ann. Conf. of the Int. FES Society,Cleveland, OH, June 16–20, 2001.

27. Kovacs, G.T.A., Storment, C.W., and Rosen, J.M., Regeneration microelectrodearray for peripheral nerve recording and stimulation, IEEE Trans. Biomed. Eng.,39(9), 893, 1992.

28. Edell, D.J., A peripheral nerve information transducer for amputees: long-term multichannel recordings from rabbit peripheral nerves, IEEE Trans.Biomed. Eng., BME-33(2), 203, 1986.

29. Rutten, W.L.C., Multi-microelectrode devices for intrafascicular use in periph-eral nerve, in Proc. 18th Ann. Int. Conf. IEEE Eng. Med. Biol. Soc., October 31– November 3, Amsterdam, 1996.



30. Banks, D.J., Ewins, D.J., Balachandran, W., and Richards, P.R., Developmentof an insertable neural signal transducer for peripheral nerve, in Proc. BESSymp. on Electrical Stimulation: Clinical Systems, April 6–7, University of Strath-clyde, Glasgow, 1995, p. S6.4.

31. Loeb, G.E., Zamin, C.J., Schulman, J.H., and Troyk, P.R., Injectable microstim-ulator for functional electrical stimulation, Med. Biol. Eng. Comput., 29, NS13,1991.

32. Ziaie, B., Nardin, M.D., Coghlan, A.R., and Najafi, K., A single-channel im-plantable microstimulator for functional neuromuscular stimulation, IEEETrans. Biomed. Eng., 44(10), 909, 1997.

33. Peachey, N.S. and Chow, A.Y., Subretinal implantation of semiconductor-based photodiodes: progress and challenges, J. Rehab. Res. Devel., 36(4), 371,1999.

34. Stelzle, M., Stett, A., Brunner, B., Graf, M., and Nisch, W., Electrical propertiesof micro-photodiode arrays for use as artificial retina implant, Biomed. Mi-crodevices, 3(2), 133, 2001.

35. Schwarz, M., Ewe, L., Hijazi, N., Hosticka, B.J., Huppertz, J., Kolnsberg, S.,Mokwa, W., and Trieu, H.K., Micro implantable visual prostheses, in Proc. 1stAnn. Int. IEEE-EMBS Special Topic Conf. on Microtechnologies in Medicine andBiology, October 12–14, Lyon, 2000, p. 461.

36. Suaning, G.J.and Lovell, N.H., CMOS neurostimulation ASIC with 100 chan-nels, scaleable output, and bi-directional radio-frequency telemetry, IEEETrans. Biomed. Eng., 48(2), 248, 2001.

37. Clements, M., Vichienchom, K., Liu, W., Hughes, C., McGucken, E., DeMarco,C., Mueller, J., Humayun, M., De Juan, E., Weiland, J., and Greenberg, R., Animplantable power and data receiver and neuro-stimulus chip for a retinalprosthesis system, in Proc. 1999 IEEE Int. Symp. on Circuits and Systems, Vol.1, 1999, p. 194.

38. Brock, K., Combining high speed and low power in 0.13 µm SoC designs,Electron. Eng. Des., May, 20, 2002.

39. Horowitz, P. and Hill, W., The Art of Electronics, 2nd ed., Cambridge UniversityPress, Cambridge, U.K., 1989.

40. Ismail, M. and Fiez, T., Eds., Analog VLSI Signal and Information Processing,McGraw-Hill, New York, 1994.

41. Geiger, R.L., Allen, P.E., and Strader, N.R., VLSI Design Techniques for Analogand Digital Circuits, McGraw-Hill, New York, 1990.

42. Soclof, S., Design and Applications of Analog Integrated Circuits, Prentice-Hall,Englewood Cliffs, NJ, 1991.

43. Jobling, D.T., Smith, J.G., and Wheal, H.V., Active microelectrode array torecord from the mammalian central nervous system in vitro, Med. Biol. Eng.Comput., 19, 553, 1981.

44. Fromherz, P., Offenhäusser, A., Vetter, T., and Weis, J., A neuron-silicon junc-tion: a Retzius cell of the Leech on an insulated-gate field-effect transistor,Science, 252, 1290, 1991.

45. Fromherz, P. and Stett, A., Silicon-neuron junction: capacitive stimulation ofan individual neuron on a silicon chip, Phys. Rev. Lett., 75(8), 1670, 1995.

46. Offenhäusser, A., Sprössler, C., Matsuzara, M., and Knoll, W., Field-effecttransistor array for monitoring electrical activity from mammalian neuronsin culture, Biosens. Bioelect., 12(8), 819, 1997.



47. Sprössler, C., Richter, D., Denyer, M., and Offenhäusser, A., Long-term re-cording system based on field-effect transistor arrays for monitoring electro-genic cells in culture, Biosens. Bioelect., 13, 613, 1998.

48. Fromherz, P., Electrical interfacing of nerve cells and semiconductor chips,Chem. Phys. Chem., 3, 276, 2002.

49. Fromherz, P., Müller, C.O., and Weis, R., Neuron transistor: electrical transferfunction measured by the patch-clamp technique, Phys. Rev. Lett., 71(24), 4079,1993.

50. Grattarola, M. and Martinioia, S., Modeling the neuron-transistor junction:from extracellular to patch recording, IEEE Trans. Biomed. Eng., 40(1), 35, 1993.

51. Ji, J., A Scaled Electrically Configurable CMOS Multichannel IntracorticalRecording Array, Ph.D. thesis, University of Michigan, Ann Arbor, 1990.

52. Bai, Q., Gingerich, M.D., and Wise, K.D., An active three-dimensional micro-electrode array for intracortical recording, in Proc. Solid-State Sensor and Ac-tuator Workshop, June 8 – 11, Hilton Head Island, SC, 1998, p. 15.

53. Edell, D.J., Toi, V.V., MvNiel, V.M., and Clark, L.D., Factors influencing thebiocompatibility of insertable silicon microshafts in cerebral cortex, IEEETrans. Biomed. Eng., 39(6), 635, 1992.

54. Drake, K.L., Wise, K.D., Farraye, J., Anderson, D.J., and BeMent, S.L., Perfor-mance of planar multisite microprobes in recording extracellular single-unitintracortical activity, IEEE Trans. Biomed. Eng., 35(9), 719, 1988.

55. Rutten, W.L. C., van Wier, H.J., Put, J.H.M., Rutgers, R., and De Vos, R.A. I.,Sensitivity, selectivity, and bioacceptance of an intraneural multi electrodestimulation device in silicon technology, Electrophysiol. Kinesiol., 135, 1998.

56. Buitenweg, J.R., Rutten, W.L.C., and Marani, E., Finite element modeling ofthe neuron-electrode interface, IEEE Eng. Med. Biol. Mag., 19(6), 46, 2000.

57. Frieswijk, T.A., Bielen, J.A., and Rutten, W.L.C., Development of a solderbump technique for contacting a three-dimensional multi electrode array, inProc. 18th Ann. Int. Conf. IEEE Eng. Med. Biol. Soc., October 31 – November 3,Amsterdam, 1996.

58. Ji, J. and Wise, K.D., An implantable CMOS circuit interface for multiplexedmicroelectrode recording arrays, IEEE J. Solid-State Circuits, 27(3), 433, 1992.

59. Peeters, E., Puers, B., and Sansen, W., A two-wire, digital output multichannelmicroprobe for recording single-unit neural activity, Sensors Actuators B, 4,217, 1991.

60. Kim, C. and Wise, K.D., A 64-site multishank CMOS low-profile neural stim-ulating probe, IEEE J. Solid-State Circuits, 31(9), 1230, 1996.

61. Tanghe, S.J. and Wise, K.D., A 16-channel CMOS neural stimulating array,IEEE J. Solid-State Circuits, 27(12), 1819, 1992.

62. Gingerich, M.D. and Wise, K.D., An active microelectrode array for multipointstimulation and recording in the central nervous system, in Proc. Transducers’99, June 7–10, Sendai, Japan, 1999, p. 280.

63. Ziaie, B., Najafi, K., and Anderson, D.J., A low-power miniature transmitterusing a low-loss silicon platform for biotelemetry, in Proc. 19th Ann. Int. Conf.IEEE Eng. Med. Biol. Soc., October 30 – November 2, Chicago, 1997, p. 2221.

64. Gatti, U. and Maloberti, F., Analog and mixed analog-digital layout, in AnalogVLSI Signal and Information Processing, Ismail, M. and Fiez, T., Eds, McGraw-Hill, New York, 1994, chap. 16.



65. Akin, T., Ziaie, B., and Najafi, K., RF telemetry powering and control ofhermetically sealed integrated sensors and actuators, in Proc. IEEE Solid-StateSensors and Actuators Workshop, 1990, p. 145.

66. Ji, J., Najafi, K., and Wise, K.D., A low-noise demultiplexing system for activemultichannel microelectrode arrays, IEEE Trans. Biomed. Eng., 38(1), 75, 1991.

67. Takahashi, K. and Matsuo, T., Integration of multi-microelectrode and inter-face circuits by silicon planar and three-dimensional fabrication technology,Sensors Actuators, 5, 89, 1984.

68. Banks, D.J., Balachandran, W., Richards, P.R., and Ewins, D., Instrumentationto evaluate neural signal recording properties of micromachined microelec-trodes inserted in invertebrate nerve, Physiol. Meas., 23, 437, 2002.

69. Bai, Q. and Wise, K.D., Single-unit neural recording with active microelec-trode arrays, IEEE Trans. Biomed. Eng., 48(8), 911, 2001.

70. Kim, K.H. and Kim, S.J., Noise characteristic design of CMOS source followervoltage amplifier for active semiconductor neural signal recording, Med. Biol.Eng. Comput., 38(4), 469, 2000.

71. Kim, K.H. and Kim, S.J., Noise performance design of CMOS preamplifierfor the active semiconductor neural probe, IEEE Trans. Biomed. Eng., 47(8),1097, 2000.

72. Dorman, M.G., Prisbe, M.A., and Meindl, J.D., A monolithic signal processorfor a neurophysiological telemetry system, IEEE J. Solid-State Circuits, SC-20(6), 1185, 1985.

73. Maher, M.P., Pine, J., Wright, J., and Tai, Y-C., The neurochip: a new multi-electrode device for stimulating and recording from cultured neurons, J. Neu-rosci. Meth., 87, 45, 1999.

74. Banks, D.J., Ewins, D.J., and Balachandran, W., The effects of high pass filter-ing extracellularly recorded action potentials predicted by computer models,Biomed. Eng. Appl. Basis Commun., 9(6), 363, 1997.

75. Kim, C. and Wise, K.D., Low-voltage electronics for the stimulation of bio-logical neural networks using fully complementary BiCMOS circuits, IEEEJ. Solid-State Circuits, 32(10), 1483, 1997.

76. Clements, M., Vichienchom, K., Liu, W., Hughes, C., McGucken, E., DeMarco,C., Mueller, J., Humayun, M., De Juan, E., Weiland, J., and Greenberg, R., Animplantable neuro-stimulator device for a retinal prosthesis, in Digest of Tech-nical Papers, IEEE Int. Solid-State Circuits Conf., 1999, p. 216.

77. Heetderks, W.J., RF powering of millimeter and submillimeter-sized neuralprosthetic implants, IEEE Trans. Biomed. Eng., 35(5), 323, 1988.

78. Shah, M.R., Philips, R.P., and Normann, R.A., A study of printed spiral coilsfor neuroprosthetic transcranial telemetry applications, IEEE Trans. Biomed.Eng., 45(7), 867, 1998.

79. Edell, D.J., Chronically implantable neural information transducers, in Proc.18th Ann. Int. Conf. IEEE Eng. Med. Biol. Soc., October 31 – November 3,Amsterdam, 1996.

80. Murakawa, K., Kobayashi, M., Nakamura, O., and Kawata, S., A wirelessnear-infrared energy system for medicinal implants, IEEE Eng. Med. Biol.Mag., Nov./Dec., 70, 1999.

81. Goto, K., Nakagawa, T., Nakamura, O., and Kawata, S., An implantable powersupply with an optically rechargeable lithium battery, IEEE Trans. Biomed.Eng., 48(7), 830, 2001.



82. Cameron, T., Loeb, G.E., Peck, R.A., Schulman, J.H., Strojnik, P., and Troyk,P.R., Micromodular implants to provide electrical stimulation of paralyzedmuscles and limbs, IEEE Trans. Biomed. Eng., 44(9), 781, 1997.

83. Schlosshauer, B., Hoff, A., Guenther, E., Zrenner, E., and Hämmerle, H., To-wards a retina prosthesis model: neurons on microphotodiode arrays in vitro,Biomed. Microdevices, 2(1), 61, 1999.

84. Stieglitz, T., Beutel, H., and Meyer, J.-U., A flexible, light-weight multichannelsieve electrode with integrated cables for interfacing regenerating peripheralnerves, Sensors Actuators A, 60, 240, 1997.

85. Rousche, P.J., Pellinen, D.S., Pivin, D.P., Williams, J.C., Vetter, R.J., and Kipke,D.R., Flexible polyimide-based intracortical electrode arrays with bioactivecapability, IEEE Trans. Biomed. Eng., 48(3), 361, 2001.

86. Akin, T., Ziaie, B., Nikles, S.A., and Najafi, K., A modular micromachinedhigh-density connector system for biomedical applications, IEEE Trans.Biomed. Eng., 46(4), 471, 1999.

87. Jones, K.E. and Normann, R.A., An advanced demultiplexing system forphysiological stimulation, IEEE Trans. Biomed. Eng., 44(12), 1211, 1997.

88. Stanley, G.B., Li, F.F., and Dan, Y., Reconstruction of natural scenes fromensemble responses in the lateral geniculate nucleus, J. Neurosci., 19(18), 8036,1999.



chapter fourteen

Molecular and nanoscale electronics

Michael C. Petty and C. Pearson

Contents

14.1 Introduction14.2 Conductive organic materials14.3 Thin-film architectures

14.3.1 Langmuir–Blodgett technique14.3.2 Self-assembly14.3.3 Electrostatic layer-by-layer deposition14.3.4 Patterning

14.4 Molecular materials for electronics14.4.1 Plastic electronics14.4.2 Organic light-emitting structures14.4.3 Chemical sensors

14.5 Molecular scale electronics14.5.1 Molecular superlattices14.5.2 Single electron devices14.5.3 DNA electronics


14.1 IntroductionThe past 30 years have witnessed the emergence of molecular electronics asan important technology for the 21st century.1,2 While modern electronicsis based largely on silicon, molecular electronics is concerned with theexploitation of organic and biological materials in electronic and opto-



electronic devices. Serious research started in the United States in the 1970swith pioneering work by Aviram, at IBM, who proposed a structure for amolecular rectifier.3 Impetus was provided by the enthusiasm of Carterworking in the U.S. Naval Research Laboratory,4 and the field was epito-mized by the elegant experiments on monolayer films undertaken by Kuhnin Göttingen.5 However, the term molecular electronics dates further back,to an ambitious project instigated by the U.S. Air Force in the 1950s in anattempt to develop the integrated circuit.6 The idea, which was radicallydifferent from other approaches to microelectronics, was to build a circuitin a solid without reproducing the individual component functions.Although analogies with the biological world are evident, the projectrequired technology that was not available at the time (and perhaps is onlynow beginning to emerge), and little progress was made.

The subject of molecular electronics, as it has matured, can broadly bedivided into two themes, as indicated in Figure 14.1, although there issubstantial overlap. The first concerns the development of electronic andopto-electronic devices using the unique macroscopic properties of organiccompounds (molecular materials for electronics). This division of molecularelectronics is already making a technological impact, the best-known exam-ple being the liquid crystal display. Other areas in which organic com-pounds are becoming increasingly important are organic light-emittingdisplays, pyroelectric plastics for infrared imaging, and chemical and bio-chemical sensors.

The second strand to molecular electronics (molecular scale electronics)recognizes the dramatic size reduction in the individual processing elements

Figure 14.1 Scope of molecular electronics.



in integrated circuits over recent years. Moore’s law — namely, the functionsper chip double every 1.5 years — will probably describe developments overat least the next decade. The semiconductor industries have produced anInternational Technology Roadmap on the future of complementary metal-oxide semiconductor (CMOS) technology.7 Figure 14.2 shows the expectedgrowth in the density of the transistors in both the microprocessor unit(MPU) and the dynamic random access memory (DRAM) of a CMOS chipover the next decade. The prediction is for a 30-nm minimum feature size(gate length) for the MPU and 1010 transistors per cm2 in the case of thememory by the year 2012. These figures are regularly updated.

14.2 Conductive organic materialsThe electronic and opto-electronic properties of silicon and gallium arsenidederive from the covalent bonds formed between the individual atoms. How-ever, semiconductive behavior is not restricted to inorganic materials. Cer-tain organic compounds possess significant electrical conductivity.8,9 Thephysical explanation can be found in the band theory for solids, one of thegreat success stories of modern physics.

Whenever two identical atoms are brought close together, the electronorbitals overlap and the energy level associated with each electron in theseparated atoms is split into two new levels, with one above and one belowthe original level. Consider, for example, the formation of a hydrogen mol-

Figure 14.2 Semiconductor Industries Association roadmap for CMOS technologyshowing the predicted growth in density of the transistors in the microprocessor unit(MPU) and the dynamic random access memory (DRAM).7



ecule from two separated atoms. When the two atoms approach each otherso that their 1s electron orbitals overlap, two new

σ-electronic orbitals areformed around the atoms, symmetric with respect to the interatomic axis.In one orbital, the bonding orbital, the electron has a lower energy than inthe isolated atom orbital and, in the other, the antibonding orbital, an electronhas a higher energy.

In an extended solid many atoms can interact and many similar splittingsof energy levels occur. For a solid containing approximately 1026 (Avogadro’snumber) atoms, each energy level splits, but the energies between these splitlevels are very small, and continuous ranges or bands of energy are formed.Two such important bands are the valence band and the conduction band,analogous to the bonding and antibonding levels of the two-atom model.The energy gap, or band gap, between them is a forbidden zone for electrons.Electrical conduction takes place by electrons moving under the influenceof an applied electric field in the conduction band and/or holes moving inthe valence band. Holes are really vacancies in a band but, for convenience,they may be regarded as positively charged carriers.

An important feature of the band model is that the electrons are delo-calized or spread over the lattice. The strength of the interaction betweenthe overlapping orbitals determines the extent of delocalization that is pos-sible for a given system. An important material parameter is the mobility ofthe charge carriers. This is defined as the carrier velocity divided by theelectric field and provides an indication of how quickly the carriers react tothe field (i.e., the frequency response of the material). The greater the degreeof electron delocalization, the larger the width of the bands (in energy terms)and the higher the mobility of the carriers within the band. For many poly-meric organic materials, the molecular orbitals responsible for bonding thecarbon atoms of the chain together are the sp3 hybridized

σ-orbitals that donot give rise to extensive overlapping. The resulting band gap is large, asthe electrons involved in the bonding are strongly localized on the carbonatoms and cannot contribute to the conduction process. This is why a simplesaturated polymer such as polyethylene, –(CH2)n–, is an electrical insulator.

A significant increase in the degree of electron delocalization may befound in unsaturated polymers (i.e., those containing double and triplecarbon–carbon bonds). If each carbon atom along the chain has only oneother atom (e.g., hydrogen) attached to it, the spare electron in a pz orbitalof the carbon atom overlaps with those of carbon atoms on either side,forming delocalized molecular orbitals of

π-symmetry. For a simple latticeof length L = Na, where N is the total number of atoms and a is the spacingbetween them, it can be shown that the total number of electron states inthe lowest energy band is exactly equal to N. This result is true for everyenergy band in the system and applies to three-dimensional lattices. Allow-ing for two spin orientations of an electron, the Pauli exclusion principlerequires that there will be room for two electrons per cell of the lattice in anenergy band. If each atom contributes one bonding electron, the valenceband will be only half filled.



From the above it might be expected that a linear polymer backboneconsisting of many strongly interacting coplanar pz orbitals, each of whichcontributes one electron to the resultant continuous

π-electron system, wouldbehave as a one-dimensional metal with a half-filled conduction band. Inchemical terms, this is a conjugated chain and may be represented by asystem of alternating single and double bonds. It turns out that, for one-dimensional systems, such a chain can more efficiently lower its energy byintroducing bond alternation (alternating short and long bonds). This limitsthe extent of electronic delocalization that can take place along the backbone.The effect is to open an energy gap in the electronic structure of the polymer.All conjugated polymers are large band-gap semiconductors, with band gapsmore than about 1.5 eV, rather than metals (the band gap in silicon is 1.1 eVat room temperature).

Figure 14.3 depicts the bond formation and electronic band structure ofa simple conjugated polymer, polyacetylene. The overlapping of the pz orbit-als to form

π bonds is shown in Figure 14.3a and the resulting band structurein Figure 14.3b. The filled

π band is the valence band, while the empty

π*

band is the conduction band. Like silicon, the conductivity of polyacetylenecan be changed by the addition of impurity atoms. However, the term dopingis a misnomer, as it tends to imply the use of minute quantities, parts permillion or less, of impurities introduced into a crystal lattice. In the case ofconductive polymers, typically 1 to 50% by weight of chemically oxidizing(electron withdrawing) or reducing (electron donating) agents are used tophysically alter the number of

π electrons on the polymer backbone, leavingoppositely charged counter ions alongside the polymer chain. These pro-cesses are redox chemistry. The doping effect can be achieved because a

πelectron can be removed (added) without destroying the

σ backbone of thepolymer so that the charged polymer remains intact. The increase in con-ductivity can be as much as 11 orders of magnitude.

Derivatives of the “basic” conductive polymers have been synthesizedto provide certain electronic features (e.g., band gap, electron affinity) andto increase their processability. Figure 14.4 shows a few examples: poly-pyrrole, polyparaphenylene, and polyphenylenevinylene. The monomerrepeat units are based on five-membered or six-membered (benzene) car-bon ring systems.

The electrical properties of organic polymers such as those describedabove are not directly comparable to those of silicon. In the latter, the three-dimensional crystallographic structure provides for extensive carrier delo-calization throughout the solid, resulting in a relatively high mobility.Electrical conduction in polymers not only requires carrier transport alongthe polymer chains but also “hopping” between these chains, which tendto lie tangled up like a plate of spaghetti. Although some improvement inthe carrier mobility can be achieved by both increasing the order of thepolymer chains and by improving the purity of the material, the chargecarrier mobilities in organic compounds are usually quite low, making itdifficult to produce very high-speed electronic computational devices that



Figure 14.3 (a) Chemical (

π) bonding in polyacteylene. (b) Electronic band structureshowing the normally empty

π* band (conduction band) and the normally filled

πband (valence band).

Figure 14.4 Chemical structures of conductive polymers: (a) polypyrrole, (b) poly-paraphenylene, and (c) polyphenylenevinylene.



are competitive with those based on silicon and gallium arsenide. None-theless, other features make organic compounds attractive for certain typesof electronic device, as indicated in the following sections.

14.3 Thin-film architecturesTechnological applications of new materials often require them to be in theform of thin films. Many thin-film-processing techniques have already beendeveloped for the fabrication of electronic and opto-electronic components.Well-established methods of organic film deposition include electrodeposition,thermal evaporation, and spinning.2 The Langmuir–Blodgett (LB) technique,self-assembly, and layer-by-layer electrostatic deposition are further means ofproducing layers of organic materials; these methods allow ultra-thin-filmassemblies of organic molecules to be engineered at the molecular level.10,11

14.3.1 Langmuir–Blodgett technique

Langmuir–Blodgett films are prepared by first depositing a small quantityof an amphiphilic compound (i.e., one containing both polar and nonpolargroups) dissolved in a volatile solvent onto the surface of purified water.11

The classical materials are long-chain fatty acids, such as n-octadecanoic acid(stearic acid), as shown in Figure 14.5. When the solvent has evaporated, the

Figure 14.5 Chemical formula for n-octadecanoic acid (stearic acid). The approximategeometrical shape and dimensions of the molecule are shown on the right.



organic molecules may be organized into a floating two-dimensional “crys-tal” by compression on the water surface. As the area available to the organicmolecules is reduced, the floating film will undergo several phase transfor-mations. These are, to a first approximation, analogous to three-dimensionalgas, liquid, and solid phases. The phase changes may readily be identifiedby monitoring the surface pressure as a function of the area occupied by themolecules in the film. This is the two-dimensional equivalent to the pressurevs. volume isotherm for a gas. Figure 14.6 shows such a plot for a hypothet-ical long-chain organic monolayer material (e.g., a long-chain fatty acid). Inthe gaseous state (G in Figure 14.6) the molecules are far enough apart onthe water surface that they exert little force on one another. As the surfacearea of the monolayer is reduced, the hydrocarbon chains will begin tointeract. The liquid state that is formed is generally called the expandedmonolayer phase (E). The hydrocarbon chains of the molecules in such afilm are in a random, rather than regular, orientation with their polar groupsin contact with the subphase. As the molecular area is progressively reduced,condensed (C) phases may appear. There may be more than one of these,and the emergence of each condensed phase can be accompanied by constantpressure regions of the isotherm, as observed in the cases of a gas condensingto a liquid and a liquid solidifying. In the condensed monolayer states, themolecules are closely packed and are oriented with their hydrocarbon chainpointing away from the water surface. The area per molecule in such a state

Figure 14.6 Surface pressure vs. area per molecule isotherm for a long-chain organiccompound showing the gaseous (G), expanded (E), and condensed (C) phases. Thesurface pressure Π and area a are in arbitrary units (a.u.).



will be similar to the cross-sectional area of the hydrocarbon chain (i.e.,

≈0.19 nm2 molecule_1).

If the surface pressure is held constant in one of the condensed phases,then the film may be transferred from the water surface onto a suitable solidsubstrate simply by raising and lowering the latter through the mono-layer–air interface. This technique, introduced by Langmuir and Blodgett,11

is illustrated in Figure 14.7. In this example, the substrate is hydrophilic andthe floating condensed monolayer (Figure 14.7a) is transferred, like a carpet,as the substrate is raised through the water (Figure 14.7b). Subsequently, a

Figure 14.7 Langmuir–Blodgett film deposition showing the transfer of amphiphilicmolecules from the surface of water onto a solid substrate.



monolayer is deposited on each traversal of the monolayer–air interface(Figure 14.7c). As shown in the figure, these stack in a head-to-head and tail-to-tail pattern (Figure 14.7d); this deposition mode is called Y-type and isthat most frequently encountered. Instances in which the floating monolayeris only transferred to the substrate as it is being inserted into the subphase,or only as it is being removed, are also observed. These deposition modesare called X-type and Z-type, respectively.

The organization of the organic molecules in LB assemblies may beinvestigated by a number of analytical techniques including x-ray and neu-tron reflection, electron diffraction, and infrared spectroscopy.11 Figure 14.8shows an example of an atomic force micrograph (AFM) of a 12-layer, n-eicosanoic acid LB film deposited onto single-crystal silicon.12 Lines of indi-vidual molecules are evident at the magnification shown.

Monolayer and multilayer films bear a striking resemblance to the nat-urally occurring biological membrane, as illustrated in Figure 14.9. The basisfor this structure is a bilayer of amphiphilic phospholipid molecules. Pro-teins, shown as large globular molecules in the figure, are partially embed-ded in and protruding from this layer. Many components of cell membranesform condensed layers at the air–water interface, and some can be assembledinto multilayer films.13 Phospholipids, such as phosphatidic acid,14 are goodexamples. However, chlorophyll-a, the green pigment in higher plants; vita-mins A, E, and K; and cholesterol can also form condensed floating mono-molecular layers. This has led to some work on LB biomimetic systems, inparticular, the development of biological sensors.15

Figure 14.8 Atomic force micrograph of a 12-layer, n-eicosanoic acid Lang-muir–Blodgett film deposited onto silicon. Unfiltered data obtained with a DigitalInstruments Nanoscope III run in contact mode.12



14.3.2 Self-assembly

Self-assembly is a much simpler process than that of LB deposition. Mono-molecular layers are formed by the immersion of an appropriate substrateinto a solution of the organic material (Figure 14.10a). The best knownexamples of self-assembled systems are organosilicon on hydroxylated sur-faces (SiO2, Al2O3, glass, etc.) and alkanethiols on gold, silver, and copper.10

However, other combinations include dialkyl sulfides on gold; dialkyl dis-ulfides on gold; alcohols and amines on platinum; and carboxylic acids onaluminum oxide and silver. The self-assembly process is driven by the inter-actions between the head group of the self-assembling molecule and thesubstrate, resulting in a strong chemical bond between the head group anda specific surface site (e.g., a covalent Si–O bond for alkyltrichlorosilanes onhydroxylated surfaces).

The combination of the self-assembly process with molecular recognitionoffers a powerful route to the development of nanoscale systems that mayhave technological applications as sensors, devices, and switches. Forinstance, the complexation of a neutral or ionic guest at one site in a moleculemay induce a change in the optical or redox properties of the system. Fig-ure 14.10b shows the proposed orientation of a derivative of the electroactivemolecule tetrathiafulvalene (TTF) on a gold surface.16 The incorporation ofthe metal-binding macrocyclic structure is to enable the molecule to functionas a metal-cation sensor. Monolayers assembled onto platinum have beenshown to exhibit electrochemical recognition to Ag+ ions.

The self-assembly process, as described above, is usually restricted tothe deposition of a single molecular layer on a solid substrate; however,chemical means can be exploited to build up multilayer organic films. Amethod pioneered by Sagiv is based on the successive absorption and

Figure 14.9 Schematic diagram showing the structure of a biological membrane.



reaction of appropriate molecules.17,18 The headgroups react with the sub-strate to give a permanent chemical attachment, and each subsequent layeris chemically attached to the one before in a very similar way to that usedin systems for supported synthesis of proteins.

14.3.3 Electrostatic layer-by-layer deposition

Another technique for building up thin films of organic molecules is drivenby the ionic attraction between opposite charges in two different polyelec-trolytes. Figure 14.11 shows a schematic diagram of this layer-by-layerassembly technique.19,20 A solid substrate with a positively charged planarsurface is immersed in a solution containing an anionic polyelectrolyte anda monolayer of polyanion is adsorbed (Figure 14.11a). Because the adsorp-tion is carried out at relatively high concentrations of the polyelectrolyte,most of the ionic groups remain exposed to the interface with the solution,

Figure 14.10 Self-assembly: (a) self-assembled monolayer film of an alkanethiol on agold-coated substrate, and (b) possible orientation of a self-assembled layer of acation-sensitive, redox-active molecule.16



thus the surface charge is reversed. After rinsing in pure water, the sub-strate is immersed in a solution containing the cationic polyelectrolyte.Again, a monolayer is adsorbed but now the original surface charge isrestored (Figure 14.11b), thus resulting in the formation of a multilayerassembly of both polymers (Figure 14.11c). It is possible to use a sensitiveoptical technique, such as surface plasmon resonance, to monitor, in situ,the growth of such electrostatically assembled films.21 The layer-by-layermethod has been used to build up layers of conductive polymers (e.g.,partially doped polyaniline and a polystyrene polyanion).22 Biocompatiblesurfaces consisting of alternate-layers of charged polysaccharides andoppositely charged synthetic polymers can also be deposited in this way.23

A related, but alternative, approach uses layer-by-layer adsorption drivenby hydrogen bonding interactions.24

14.3.4 Patterning

Patterning technologies are essential in the fabrication of most microlectronicdevice structures. Planar components are normally patterned using photo-lithography. Here, a surface is first covered with a light-sensitive photoresist,which is exposed to ultraviolet light through a contact mask. Either theexposed photoresist (positive resist) or the unexposed regions (negativeresist) can then be washed away to leave a positive or negative image of themask on the surface. This approach is routinely used in the fabrication ofdevices based on inorganic semiconductors. However, difficulties can beencountered when used with organic films, as the photoresists themselvesare based on organic compounds.

Figure 14.11 Schematic representation of the assembly of a multilayer architectureby consecutive adsorption of anionic and cationic polyelectrolytes.19



Brittain et al.25 describe a series of “soft” lithographic methods thatmay be better suited to the patterning of organic layers. Pouring a liquidpolymer, such as polydimethylsiloxane (PDMS) onto a “master” made fromsilicon forms a pattern-transfer element, as shown in Figure 14.12. Thepolymer is allowed to cure to form an elastomer, which can then beremoved from the master (Figure 14.12a). This replica can subsequently beused as a stamp to transfer chemical ink, such as a solution of an alkaneth-iol, to a surface (Figure 14.12b).

Scanning microscopy methods offer a powerful means of manipulatingmolecules. Careful control of an AFM tip can allow patterns to be drawn inan organic film.26 Such techniques can also be used to reposition molecules,such as the fullerene C60, on surfaces and to break up an individual mole-cule.27 A further approach that has recently been developed at NorthwesternUniversity is called dip-pen nanolithography (DPN).28 This technique, illus-trated in the Figure 14.13, is able to deliver organic molecules in a positiveprinting mode. An AFM tip is used to “write” alkanethiols on a gold thinfilm in a manner analogous to that of a fountain pen. Molecules flow fromthe AFM tip to a solid substrate (“paper”) via capillary transport, makingDPN a potentially useful tool for assembling nanoscale devices.

The chemisorption of the “ink” is the driving force that moves the inkfrom the AFM tip through the water to the substrate as the tip is scannedacross this surface. Adjusting the scan rate and relative humidity cancontrol line widths. The current line-width resolution is 15 nm and spatialresolution is less than 10 nm. These parameters are limited by the tip radiiof curvature associated with conventional AFM tips but in principle can

Figure 14.12 Schematic representation of an example of soft lithography: (a) forma-tion of an elastomeric stamp, and (b) transfer of chemical “ink” to a surface.25



be decreased through improvements in tip fabrication technology. DPN isa direct-write nanolithographic process where one can pattern and imagewith the same tool.

Recent developments of DPN have included an overwriting capabilitythat allows one nanostructure to be generated and the areas surroundingthat nanostructure to be filled with a second type of ink.29 Perhaps thegreatest limitation in using scanning probe methodologies for ultra-high-resolution nanolithography over large areas derives from the serial natureof most techniques; however, an eight-pen nanoplotter capable of doingparallel lithography has been reported.30 The DPN method has also beenused to deposit magnetic nanostructures31 and arrays of protein molecules.32

There is also an intriguing link between the DPN process and LBdeposition. For example, if the DPN experimental arrangement shown inFigure 14.13 is turned by 90

°, a nanometer-scale LB trough, similar to thosereported that use a moving subphase to compress the monolayer film isevident.33 As the amphiphilic molecules flow down the AFM tip, supportedby a thin layer of water, their surface pressure will rise. The resultingcondensed monomolecular film is then transferred to a solid substrate;however, in the case of DPN, the substrate is stationary while the “nano-trough” moves in relation to it.

14.4 Molecular materials for electronics14.4.1 Plastic electronics

Since the discovery of semiconducting behavior in organic materials, con-siderable research effort has been aimed at exploiting these properties inelectronic and opto-electronic devices. Organic semiconductors can havesignificant advantages over their inorganic counterparts. For example, thinlayers of polymers can easily be made by low-cost methods such as spincoating. High-temperature deposition from vapor reactants is generallyneeded for inorganic semiconductors. Synthetic organic chemistry also offersthe possibility of designing new materials with different bandgaps. As notedin Section 14.2, the mobilities of the charge carriers in organic field effect

Figure 14.13 Dip-pen patterning technique showing the transfer of an “ink” onto“paper” using the tip of an atomic force microscope (AFM).28



transistors are low. Nevertheless, the simple fabrication techniques for poly-mers have attracted several companies to work on polymer transistor appli-cations such as data storage and thin-film device arrays to address liquidcrystal displays.34

Semiconducting organic films have been used similarly to inorganicsemiconductors (e.g., Si, GaAs) in metal/semiconductor/metal structures.Perhaps the simplest example is that of a diode. Here, the semiconductor issandwiched between metals of different work functions. In the ideal case,an n-type semiconductor should make an ohmic contact to a low workfunction metal and a rectifying Schottky barrier to a high work functionmetal.35 An early example is that of a phthalocyanine LB film sandwichedbetween aluminum and indium–tin–oxide electrodes.36

Of particular interest is the possibility of observing molecular rectifi-cation using monolayer or multilayer films. This follows the early predic-tion of Aviram and Ratner (noted in the Introduction) that an asymmetricorganic molecule containing a donor and an acceptor group separated bya short σ-bonded bridge, allowing quantum mechanical tunneling, shouldexhibit diode characteristics.3 Many attempts have been made to demon-strate this effect in the laboratory, particularly in LB film systems.37,38 Asym-metric current vs. voltage behavior has certainly been recorded for manyLB film metal/insulator/metal structures, although some of these resultsare open to several interpretations as a result of the asymmetry of theelectrode configuration.

Organic thin films have also been used as the semiconducting layer infield effect transistor (FET) devices.39,40 These are three-terminal structures;a voltage applied to a metallic gate affects an electric current flowing betweensource and drain electrodes. For transistor operation, the charge must beeasily injected from the source electrode into the organic semiconductor andthe carrier mobility should be high enough to allow useful quantities ofsource–drain current to flow. The organic semiconductor and other materialswith which it is in contact must also withstand the operating conditionswithout thermal, electrochemical, or photochemical degradation. Two per-formance parameters to be optimized in organic field effect transistors arethe field effect mobility and on/off ratio.40

Figure 14.14 shows a schematic diagram for the possible structure of anorganic thin-film FET. In this arrangement, the organic film is deposited inthe final stages of FET fabrication. It is therefore not necessary for this layerto withstand any post deposition, chemical, and thermal processing. Fig-ure 14.15 shows how a series of transistor devices can be made on the samesilicon substrate incorporating different thicknesses of LB films. The siliconserves as the gate electrode, while silicon dioxide forms the insulator. Exper-imental data for such a device, using an organometallic complex as thesemiconductive layer, are shown in Figure 14.16.41,42 The graph shows thedependence of the saturated source–drain current vs. gate bias voltage fora device incorporating a film consisting of 59 LB layers of the iodine-dopedcomplex on top of the interdigitated source and drain electrodes.42 The slope



Figure 14.14 Schematic diagram of a field effect transistor (FET) structure.

Figure 14.15 Plan view of organic transistors on a silicon substrate. The structure isthat shown in Figure 14.14. The gate is silicon. Silicon dioxide forms the gate insulator,while the semiconductor is a Langmuir–Blodgett film. A stepped structure providesdifferent thicknesses of semiconductive film.42

Silicon substrate

Gold

Silicon dioxide

LB filmsteppedstructure

5x5 array of interdigitated electrode structures

Silicon dioxide etched awayto reveal silicon substrate

Gate contacts



of the straight line gives a carrier mobility of 0.3 cm2 V–1 sec,–1 a relativelyhigh figure for an organic transistor device.

The operating characteristics of organic transistors have improved mark-edly over recent years. This has been brought about by both improvementsin the material synthesis and in the thin-film processing techniques.43–48 State-of-the-art devices possess characteristics similar to those of devices preparedfrom hydrogenated amorphous silicon, with mobilities around 1 cm2 V–1 sec–1

and on/off ratios greater than 106. The use of ambipolar thin-film devicesshould lead to a significant simplification of complementary logic circuits.Thin-film transistors based on organic semiconductors are likely to form keycomponents of plastic circuitry for use as display drivers in portable com-puters and pagers, and as memory elements in transaction cards and iden-tification tags.

14.4.2 Organic light-emitting structures

Reports of light emission from organic materials on the application of anelectric field (electroluminescence) have been around for many years.However, there has been an upsurge in interest following the initial reportof organic light-emitting devices (OLEDs) incorporating the conjugatedpolymer poly(p-phenylene vinylene) (PPV).49 The simplest OLED is anorganic semiconductor sandwiched between electrodes of high and lowwork function. On application of a voltage, electrons are injected fromone of the electrodes and holes from the other; recombination of thesecarriers then results in the emission of light. Electron- and hole-transport-ing films may be deposited between the cathode and the emitting layerand the anode and the emitting layer, respectively, to improve and balancethe injection of carriers. Figure 14.17 shows the structure of an OLED that

Figure 14.16 (Saturated drain-source current)0.5 vs. gate bias voltage for a thin-filmtransistor incorporating 59 Langmuir–Blodgett layers of an iodine-doped chargetransfer complex on top of interdigitated source and drain electrodes.41,42

0

0.01

0.02

0.03

0.04

10 20 30

Dra

in c

urre

nt0.

5 (A

0.5 )

40 50Gate bias (V)



uses a hole transport layer sandwiched between an indium–tin–oxide(ITO) anode and a light-emitting layer. Work is focused on the use of low-molecular-weight organic molecules and polymers and industrial interestin the application of such materials to various display technologies isconsiderable.50

Many techniques have been used in attempts to optimize the perfor-mance of OLEDs. For example, an inorganic insulating layer, such as LiF,may be inserted between the cathode and the emissive material.50 The ITOanode electrode can be treated (e.g., with an oxygen plasma51) to reduce theturn-on voltage. Bilayer anodes such as ITO/polyaniline,52 ITO/poly-(3,4-ethylene dioxythiophene),53 and ITO/phthalocyanine54 have also beenshown to be beneficial.

The thin polymeric or oligomeric layers required for OLEDs are usuallydeposited by spin coating or thermal evaporation. The LB technique andlayer-by-layer self-assembly offer alternative methods for building up ultra-thin organic films of nanometer dimensions. These approaches to thin-filmdeposition allow the thickness requirements of OLEDs to be fully exploredin the laboratory. They may also offer certain advantages for the fabricationof devices containing ultrathin organic films and/or alternate-layer films. Ina similar manner to the use of LiF noted above, improvements in deviceefficiency may be achieved by the insertion of an LB film “spacer” betweenthe emissive film and the cathode.55 The relatively ordered nature of LBarrays of molecules may be exploited in light-emitting structures. For exam-ple, the preferential alignment of the molecules can result in polarized lightemission.56,57 Langmuir–Blodgett deposition can also be used to control the

Figure 14.17 Organic light-emitting device structure incorporating a light-emittinglayer and a hole transport layer sandwiched between an indium–tin–oxide (ITO)anode and a metallic cathode.



positions of luminescent species within metal mirror microcavities, allowingthe optical mode structure of the cavities to be probed by both the emissionand absorption of the chromophore.58

14.4.3 Chemical sensors

The development of effective devices for the identification and quantificationof chemical and biochemical substances for process control and environmen-tal monitoring is a growing need.59,60 Many sensors do not possess the spec-ifications to conform to existing or forthcoming legislation; some systemsare too bulky or expensive for use in the field. Inorganic materials such asthe oxides of tin and zinc have traditionally been favored as the sensingelement;61 however, one disadvantage of sensors based on metallic oxides isthat they usually have to be operated at elevated temperatures, limiting someapplications. As an alternative, there has been considerable interest in tryingto exploit the properties of organic materials. Many such substances (inparticular, phthalocyanine derivatives) are known to exhibit a high sensitiv-ity to gases.62 Lessons can also be taken from the biological world; onehousehold carbon monoxide detector currently being marketed is designedto simulate the reaction between CO and hemoglobin. A significant advan-tage of organic compounds is that their sensitivity and selectivity can betailored to a particular application by modifications to their chemical struc-ture. Moreover, thin-film technologies, such as self-assembly or layer-by-layer electrostatic deposition, enable ultra-thin layers of organic materials tobe engineered at the molecular level.63

There are many physical principles upon which sensing systems mightbe based; changes in electrical resistance (chemiresistors), refractive index(fiberoptic sensors), and mass (quartz microbalance) have all been exploitedin chemical sensing. The main challenges in the development of new sensorsare in the production of inexpensive, reproducible, and reliable devices withadequate sensitivities and selectivities.

While many sensing devices may show adequate sensitivities, the selec-tivity can be poor. A semiconductive polymer may show a similar changein electrical resistance to a range of oxidizing (reducing) gases. To get aroundthis difficulty, one approach that is being embraced enthusiastically byresearchers is to use an array of sensing elements, rather than a single device.This is the method favored by nature. The human olfactory system, depictedin Figure 14.18a, has many receptors cells (sensors), which are individuallynonspecific; signals from these are fed to the brain via a network of primary(glomeruli layer) and secondary (mitral layer) neurons for processing. It isgenerally believed that the selectivity of the olfactory system is a result of ahigh degree of parallel processing in the neural architecture. Artificial neuralnetworks (Figure 14.18b) can, to some extent, emulate the connectivity ofthe olfactory neurons.64 The electronic nose is an attempt to mimic the humanolfactory system, and several companies now market such equipment.60,65

Individual sensors can be based on polymer films. Each element is treated



in a slightly different way during deposition so that it responds uniquely onexposure to a particular gas or vapor. The pattern of resistance changes inthe sensor array can then be used to fingerprint the vapor.

An alternative technique to using an array of sensing elements is tomake multiple measurements on one sample. For example, discriminationbetween different vapors may be achieved by monitoring the complexelectrical admittance of a sensing layer at several frequencies (admittancespectroscopy). Figure 14.19 shows the frequency behavior, at room temper-ature over the range of 10–1 to 103 kHz, of the capacitance of an LB film ofa coordination polymer formed by the reaction of the bifunctionalamphiphilic ligand, 5,5′methylenebis (N-hexadecylsalicylideneamine) andcopper ions in an interfacial reaction at the water surface.66,67 The measure-ments were undertaken in both nitrogen gas and ethanol vapor (3.3%). Theinset shows the transient behavior measured at a fixed frequency (1 kHz)when the ethanol vapor was turned on and off. On exposure to othervapors, similar transient responses were noted. The percentage changes inthe capacitance and conductance, normalized to the vapor concentrations,are shown in Figure 14.20.

It is evident that the capacitance increases when the device is exposedto ethanol and acetonitrile, with a greater fractional increase for the aceto-nitrile. This is consistent with the strongly polar nature of the latter solvent(dipole moment for acetonitrile = 3.92 Debye compared to 1.69 Debye for

Figure 14.18 (a) Human olfactory system. (b) Artificial neural network.

Olfactory receptor cell

Odorin

Odorin

Glomerulilayer

Mitrallayer

Granularcell layer

Brain

Primaryneurons

Secondaryneurons

Sensorarray

Identifiedodor

ArtificialNeural

Network

(a)

(b)



ethanol). A decrease of capacitance is observed with the nonpolar benzene,almost certainly associated with the swelling of the film. From a practicalsensing viewpoint, it is clear that monitoring the admittance at four frequen-cies provides a means of discrimination between the vapors studied.

14.5 Molecular scale electronicsThe “bottom-up” approach to molecular electronics offers many intriguingprospects for manipulating materials on the nanometer scale, thereby pro-viding opportunities to build up novel architectures with predetermined andunique physical and/or chemical properties. Two relatively simple examplesare described below: the incorporation of an electric polarization into amultilayer array and the use of conductive nanoparticles to realize singleelectron devices. The possible exploitation of the electronic properties ofDNA is also discussed.

14.5.1 Molecular superlattices

It is widely recognized that pyroelectric materials have considerable advan-tages over narrow bandgap semiconductors, such as mercury cadmium tel-luride (CMT), as detectors of infrared radiation. In addition to their broaderspectral sensitivity, pyroelectric detectors possess the advantage of efficientoperation at ambient temperatures, obviating the need for expensive coolingsystems that are required for their CMT counterparts.68

For a material to be pyroelectric, it must possess a noncentrosymmetriccrystal structure with a unique polar axis.69 Furthermore, if the material isto form the basis of an efficient pyroelectric detector, it must be fabricated

Figure 14.19 Capacitance vs. frequency for a coordination polymer Lang-muir–Blodgett film in nitrogen and exposed to ethanol vapor (3.3%). The inset showsthe transient behavior measured at a fixed frequency (1 kHz) when the ethanol vaporwas turned on and off (indicated by the arrows).66

10 -1 1 10 10 2 10 3190

200

210

220

230

nitrogen ethanol

Cap

acita

nce

[pF

]

Frequency [kHz]

60200

205

210

215

time [min]

Cap

acita

nce

[pF

]

40200



in thin-film form. Conventionally, two broad approaches have been taken toproducing pyroelectric thin films: the first is to grow a single crystal of amaterial such as triglycine sulfate, which belongs to one of the polar spacegroups, and to thin the crystal by slicing, mechanical grinding, or etching;the second, and more satisfactory method is to grow a thin polycrystalline,nonpolar film and to render it pyroelectric by applying a large electric field(“poling”). This is particularly applicable to ceramics, which can be depos-ited by radio frequency sputtering, and to polymers, which can be spin-coated onto a suitable substrate.

Several advantages make LB films particularly attractive candidates forpyroelectric materials. The most important is that the sequential depositionof single monolayers enables the symmetry of the film to be preciselydefined; in particular, layers of different materials can be built up to producea highly polar structure. Second, the polarization of an LB film is “frozen-in” during deposition, and it is therefore not necessary to subject the film toa poling process. The third advantage is that the LB technique usesamphiphilic organic materials which possess low permittivities, ε, and the

Figure 14.20 Average changes in (a) the capacitance, ∆C, and (b) the conductance,∆G, for a coordination polymer Langmuir–Blodgett film, measured at four frequen-cies and for three different vapors.66

1 kHz 10 kHz 100 kHz 1000 kHz

-1

0

1

2

a)

<

C >

[

%]

benzene ethanol acetonitrile

1 kHz 10 kHz 100 kHz 1000 kHz0

2

4

6

8

10

12

14

16

18b)

<

G >

[

%]



figure of merit for voltage responsivity p/ε (where p is the pyroelectriccoefficient) is expected to be large. Finally, the LB method enables the prep-aration of much thinner films than are usually attainable by more conven-tional techniques.

Figure 14.21 illustrates the principle of a superlattice consisting of acidand amine molecules whose dipole moments are in opposite senses but whendeposited in Y-type LB film form are aligned in the same direction. Infraredstudies have indicated that the deposition results in a proton transfer fromthe acid head group to that of the amine, giving rise to an overall polarizationcomponent perpendicular to the multilayer plane.70

Many LB materials deposited as alternate-layer films or as X- or Z-typelayers exhibit such behavior, including polymeric compounds and phos-pholipid materials.71 The latter compounds can possess large dipolemoments associated with their head groups. The pyroelectric coefficientsof multilayer acid/amine LB films can be about 10 µC m–2 K–1 and dependon the thermal expansion coefficient of the substrate, indicating the pres-ence of a significant secondary contribution (i.e., as a result of the piezo-electric effect) to the measured pyroelectric response.68,70 Although thesepyroelectric coefficients are still less (by about a factor of three) than thosemeasured for polyvinylidene fluoride, a well-known pyroelectric polymer,the dielectric constants are also less, providing comparable figures of meritfor infrared detection devices. Alternate-layer LB structures can also exhibita significant second-order nonlinear optical response (e.g., second-har-monic generation).11

Figure 14.21 Schematic diagram of an organic superlattice formed by the alternate-layer Langmuir–Blodgett deposition of a long-chain acid and a long-chain amine.The dipole moments associated with the polar head groups are shown on the right.68



14.5.2 Single electron devices

Progress down the Moore’s law curve (Figure 14.2) has enabled the designof electronic structures on a scale where quantum mechanical effects becomeimportant. The current flow in such devices can be determined by tunnelingthrough energy barriers and may also exploit the fact that charge is quantizedin units of e (1.6 × 10–19 C). Consequently, there has been much interestfocused on single-electron devices, in which the addition or subtraction ofa small number of electrons to or from an electrode can be controlled withone-electron precision.72,73

A single-electron tunneling device is a structure based on the principleof Coulomb blockade, where the number of electrons on an island (semicon-ducting or metallic dot) is controlled by a capacitatively coupled gate, asdepicted in Figure 14.22a.

The associated change in Coulomb energy on addition of one electroniccharge is conveniently expressed in terms of the capacitance, C, of the island.The extra charge changes the electrostatic potential by the charging energy,e2/C. This charging energy becomes important when it exceeds the thermalenergy kBT, where kB is the Boltzmann constant (1.38 × 10–23 JK–1) and T is theabsolute temperature.

A second condition is that the barrier between the electron reservoir andthe island is sufficiently opaque that the electrons are either located on theisland or in the reservoir. This means that quantum fluctuations in the num-ber due to tunneling through the barriers is much less than one. This require-ment translates to a lower value for the tunnel resistance, Rt, of the barrier,which should be much larger than the resistance quantum, h/e2 = 25.813 kΩin order for the energy uncertainty to be much smaller than the chargingenergy. The conditions to be fulfilled are:73

The first criterion can be met by weakly coupling the island to the electronreservoir. The second condition is satisfied by making the island very small.Room-temperature operation requires structures of less than 10 nm, muchsmaller than present lithography resolutions and difficult to achieve.

Figure 14.22b shows the operation of the tunneling device. By applying apositive gate voltage, an electron is transferred to the dot at the critical gatevoltage. At this moment, the potential in the dot region is decreased and blocksthe transfer of other electrons (Coulomb blockade). If the nanoparti-cles/gate/reservoir form part of an FET, then this trapping of an electron inthe dot shifts the threshold voltage of the transistor. Therefore, by sensing thecurrent difference between the two states, the stored information can be read.

Rhe

eC

k T

t

B

>>

>>

2

2

and



Many approaches have been used to form the size of nanoparticlenecessary for operation of such single-electron devices at room tempera-ture. For example, iron oxide nanoparticles can be synthesized by thereduction of ferric chloride.74 Langmuir–Blodgett layers of the cadmiumsalt of a long-chain fatty acid can be converted to arrays of CdS nanoclus-ters by exposure to H2S gas.75,76 Nanoparticles consisting of metallic parti-cles capped with a short-chain oligomer can be manipulated directly bythe LB method. Figure 14.23 shows a transmission electron micrograph ofsuch particles deposited onto a carbon-coated microscope grid.77 The indi-vidual particles, with a mean diameter of approximately 8 nm, are orga-nized in a close-packed arrangement.

Figure 14.22 (a) Principle of Coulomb blockade; transfer of single electrons from anelectrode (reservoir) to a conductive dot (island). (b) Schematic energy diagram fora single-electron memory showing the writing of one bit of data.72



14.5.3 DNA electronics

The study of the electronic behavior of organic compounds has led somescientists to work on the electrical properties of biological materials. Deox-yribonucleic acid (DNA) is arguably the most significant molecule in nature.Reports of the electronic properties of DNA have already generated contro-versy in the literature.78–82 According to some, DNA is a molecular wire ofvery small resistance; others, however, find that DNA behaves as an insula-tor. These seemingly contradictory findings can probably be explained bythe different DNA sequences and experimental conditions used to monitorthe conductivity.79

The DNA strand in the double-helix arrangement consists of a longpolymer backbone comprising repeating sugar molecules and phosphategroups. Each sugar group is attached to one of four bases, guanine (G),cytosine (C), adenine (A), and thymine (T). The chemical bonding is suchthat an A base only ever pairs with a T base, while a G always pairs with aC. Some of the electron orbitals belonging to the bases overlap quite wellwith each other along the long axis of the DNA. This provides a path forelectron transfer along the molecule, in a similar fashion to the one-dimen-sional conduction seen in conjugated polymers (see Section 14.2).

Theoretical and experimental work now suggests that a hole (i.e., apositive charge) is more stable on a G–C base pair than on an A–T base pair.79

The energy difference between these two pairs is substantially larger thanthe thermal energy of the charge carrier. Under these conditions, a hole willlocalize on a particular G–C base pair. Because the A–T base pairs have ahigher energy, they act as a barrier to hole transfer. However, if the distancebetween two G–C base pairs is small enough, the hole can tunnel quantum

Figure 14.23 Transmission electron micrograph of a Langmuir–Blodgett layer ofgold nanoparticles deposited onto a carbon-coated grid.77



mechanically from one pair to the next. In this way, charge carriers are ableto shuttle along a single DNA molecule over a distance of a few nanometers.

“DNA chips” exploit the fact that short strands of DNA will bind to othersegments of DNA that have complementary sequences and can therefore beused to probe whether certain genetic codes are present in a given specimenof DNA. Microfabricated chips with many parallel DNA probes are becomingwidespread in analytical and medical applications. Currently, the chips areread out optically, but further miniaturization might require new read-outschemes, possibly involving the electron-transfer properties of DNA.

Computations by chemical or biological reactions overcome the problemof parallelism and interconnections in a classical system. If a string of DNAcan be put together in the proper sequence, it can be used to solve combi-national problems. The calculations are performed in test tubes filled withstrands of DNA. Gene sequencing is used to obtain the result. For example,Adleman83 calculated the “traveling salesman” problem to demonstrate thecapabilities of DNA computing. DNA computing on parallel problemspotentially provides 1014 MIPS (millions of instructions per second) and usesless energy and space than conventional supercomputers. While CMOSsupercomputers operate 109 operations per Joule, a DNA computer couldperform about 1019 operations per Joule. Data could potentially be stored onDNA in a density of approximately 1 bit per nm3 while existing storagemedia such as DRAMs require 1012 nm3 to store 1 bit.

14.6 ConclusionsOrganic compounds possess a wide range of fascinating physical and chem-ical properties that make them attractive candidates for exploitation in elec-tronic and opto-electronic devices. It is not, however, expected that thesematerials will displace silicon in the foreseeable future as the dominantmaterial for fast signal processing. It is far more likely that organic materialswill find uses in other niche areas of electronics, where silicon and otherinorganic semiconductors cannot compete. Examples already exist, such asliquid crystal displays and certain chemical sensors. Organic light emittingstructures are likely to make a major impact in the market place over thenext ten years.

For commercialization, organic materials are usually required in theform of thin films. Simple processing methods, such as spin-coating andthermal evaporation, are already available. A number of other thin-filmtechnologies offer the means to manipulate arrays of organic molecules (i.e.,molecular engineering). Techniques such as Langmuir–Blodgett film depo-sition, self-assembly, and layer-by-layer electrostatic deposition allowintriguing molecular architectures to be built up on solid surfaces. Quantummechanical effects can control the behavior of devices based on such assem-blies of molecules.

Over the first decades of the 21st century, classical CMOS technologywill come up against a number of technological challenges. The bottom-up



approach to molecular electronics provides an alternative and attractive wayforward and, as such, it is currently an area of exciting interdisciplinaryactivity. Analogies with the living world are seductive and much can belearnt from nature (for example, neural networks for chemical sensing).However, it is rather too early to tell whether sophisticated computing struc-tures mimicking the human brain can be produced. It is probably not appro-priate to emulate individual inorganic devices (e.g., Si MOSFETs) usingorganic, or even biological, compounds, and then to connect these togetherusing the techniques developed by the microelectronics industry. The “wir-ing” problems could be insurmountable. Living systems assemble them-selves from molecules and are extremely energetically efficient when com-pared with manmade computational devices. More radical approaches tomaterials fabrication and device design (e.g., exploiting self-organization)are almost certainly required if molecular-scale electronics is to be realized.

References1. Petty, M.C., Bryce, M.R., and Bloor, D., Eds., An Introduction to Molecular

Electronics, Edward Arnold, London, 1995.2. Richardson, T.H., Ed., Functional Organic and Polymeric Materials, John Wiley

& Sons, Chichester, 2000.3. Aviram, A. and Ratner, M.A., Chem. Phys. Lett., 29, 277, 1974.4. Carter, F.L., J. Vac. Sci. Technol. B, 1, 959, 1983.5. Kuhn, H., in Molecular Electronics, Hong, F.T., Ed., Plenum Press, New York,

1989.6. Braun, E. and MacDonald, S., Revolution in Miniature, Cambridge University

Press, Cambridge, U.K., 1978.7. Semiconductor Industry Association Roadmap, http://public.itrs.net/.8. Ferraro, J.R. and Williams, J.M., Introduction to Synthetic Electrical Conductors,

Academic Press, Orlando, FL, 1987.9. Roth, S., One-Dimensional Metals, VCH, Weinheim, 1995.

10. Ulman, A., Ultrathin Organic Films, Academic Press, San Diego, CA, 1991.11. Petty, M.C., Langmuir-Blodgett Films, Cambridge University Press, Cambridge,

U.K., 1996.12. Evanson, S.A., Badyal, J.P.S., Pearson, C., and Petty, M.C., J. Phys. Chem., 100,

11672, 1996.13. Swart, R.M., in Langmuir-Blodgett Films, Roberts, G.G., Ed., Plenum Press,

New York, 1990.14. Cui, D.F., Howarth, V.A., Petty, M.C., Ancelin, H., and Yarwood, J., Thin Solid

Films, 192, 391, 1990.15. Nicolini, C., Ed., Molecular Bioelectronics, World Scientific, Singapore, 1996.16. Moore, A.J., Goldenberg, L.M., Bryce, M.R., Petty, M.C., Monkman, A.P.,

Marenco, C., Yarwood, J., Joyce, M.J., and Port, S.N., Adv. Mater., 10, 395, 1998.17. Netzer, L. and Sagiv, J., J. Am. Chem. Soc., 105, 674, 1983.18. Netzer, L., Iscovici, R., and Sagiv, J., Thin Solid Films, 99, 235, 1983.19. Decher, G., Hong, J.D., and Schmitt, J., Thin Solid Films, 210/211, 831, 1992.20. Decher, G., Lvov, Y., and Schmitt, J., Thin Solid Films, 244, 772, 1994.21. Pearson, C., Nagel, J., and Petty, M.C., J. Phys. D: Appl. Phys., 34, 285, 2001.



http://public.itrs.net/

22. Cheung, J.H., Stockton, W.B., and Rubner, M.F., Macromolecules, 30, 2712, 1997.23. Lvov, Y., Onda, M., Ariga, K., and Kunitake, T., J. Biomater. Sci., Polym. Ed., 9,

345, 1998.24. Stockton, W.B. and Rubner, M.F., Macromolecules, 30, 2717, 1997.25. Brittain, S., Paul, K., Zhao, X.-M., and Whitesides, G., Phys. World, 11, 31, 1998.26. Chi, L.F., Eng, L.M., Graf, K. and Fuchs, H., Langmuir, 8, 2255, 1992.27. Gimzewski, J., Phys. World, 11, 25, 1998.28. Piner, R.D., Zhu, J., Xu, F., Hong, S., and Mirkin, C.A., Science, 283, 661, 1999.29. Hong, S., Zhu, J., and Mirkin, C.A., Science, 286, 523, 1999.30. Hong, S. and Mirkin, C.A., 288, 1808, 2000.31. Liu, X.G., Fu, L., Hong, S.H., Dravid, V.P., and Mirkin, C.A., Adv. Mater., 14,

231, 2002.32. Lee, K.B., Park, S.J., Mirkin, C.A., Smith, J.C., and Mrksich, M., Science, 295,

1702, 2002.33. Lu, Z., Qian, F., Zhu, Y., Yang, X., and Wei, Y., Thin Solid Films, 243, 371, 1994.34. May, P., Phys. World, 8, 52, 1995.35. Rhoderick, E.H., Metal-Semiconductor Contacts, Clarendon Press, Oxford, 1978.36. Hua, Y.L., Petty, M.C., Roberts, G.G., Ahmad, M.M., Hanack, M., and Rein,

M., Thin Solid Films, 149, 161, 1987.37. Martin, A.S., Sambles, J.R., and Ashwell, G.J., Phys. Rev. Lett., 70, 218, 1993.38. Ashwell, G.J. and Gandolfo, D.S., J. Mater. Chem., 11, 246, 2001.39. Horowitz, G., Adv. Mater., 2, 286, 1990.40. Katz, H.E., J. Mater. Chem., 7 369, 1997.41. Pearson, C., Gibson, J.E., Moore, A.J., Bryce, M.R. and Petty, M.C., Electron.

Lett., 29, 1377, 1993.42. Pearson, C., Langmuir–Blodgett Films of Organic Charge-Transfer Complex-

es, Ph.D. thesis, University of Durham, U.K., 1996.43. Laquindanum, J.G., Katz, H.E., Lovinger, A.J., and Dodabalapur, A., Chem.

Mater., 8, 2542, 1996.44. Bao, Z., Lovinger, A.J., and Dodabalapur, A., Adv. Mater., 9, 42, 1997.45. Dimitrakopoulos, C.D., Furman, B.K., Graham, T., Hedge, S., and Purushotha-

man, S., Synth. Met., 92, 47, 1998.46. Brown, A.R., Pomp, A., Hart, C.M., and de Leeuw, D.M., Science, 270, 972,

1995.47. Dimitrakopoulos, C.D., Purushothaman, S., Kymissis, J., Callegari, A., and

Shaw, J.M., Science, 283, 822, 1999.48. Sirringhaus, H., Kawase, T., Friend, R.H., Shimoda, T., Inbasekaran, M., Wu,

W., and Woo, E.P., Science, 290, 2123, 2000.49. Burroughes, J.H., Bradley, D.D.C., Brown, A.R., Marks, R.N., Mackay, K.,

Friend, R.H., Burns, P.L., and Holmes, A.B., Nature, 347, 359, 1990.50. Hudson, A.J. and Weaver, M.S., in Functional Organic and Polymeric Materials,

Richardson, T.H., Ed., John Wiley & Sons, Chichester, 2000, p. 365.51. Kim, J.S., Granstrn, M., Friend, R.H., Johansson, N., Salaneck, W.R., Daik,

R.,Feast, W.J., and Cacialli, F., J. Appl. Phys., 84, 6859, 1998.52. Karg, S., Scott, J.C., Salem, J.R., and Angelopoulos, M., Synth. Meth., 80, 111,

1996.53. Brown, T.M., Kim, J.S., Friend, R.H., Cacialli, F., Daik, R., and Feast, W.J., Appl.

Phys. Lett., 75, 1679, 1999.54. Baldo, M.A., O’Brien, D.F., You, Y., Shoustikov, A., Sibley, S., Thompson, M.E.,

and Forrest, S.R., Nature, 395, 151,1998.



55. Jung, G.Y., Pearson, C., Horsburgh, L.E., Samuel, I.D.W., Monkman, A.P., andPetty, M.C., J. Phys. D: Appl. Phys., 33, 1029, 2000.

56. Sterbacka, R., Juka, G., Arlauskas, K., Pal, A.J., Kllman, K.-M., and Stubb, H.,J. Appl. Phys., 84, 3359, 1998.

57. Cimrov, V., Remmers, M., Neher, D. and Wegner, G., Adv. Mater., 8, 146, 1996.58. Burns, S.E., Pfeffer, N., Grηner, J., Neher, D., and Friend, R.H., Synth. Meth.,

84, 887, 1997.59. Janata, J., Principles of Chemical Sensors, Plenum Press, New York, 1989.60. Gardner, J.W., Microsensors, John Wiley & Sons, Chichester, 1994.61. Moseley, P.T. and Crocker, A.J., Sensor Materials, IOP Publishing, Bristol, 1996.62. Snow, A.S. and Barger, W.R., in Phthalocyanines: Properties and Applications,

Leznoff, C.C. and Lever, A.B.P., Eds., VCH Publishers, New York, 1989, p. 342.63. Petty, M.C. and Casalini, R., Eng. Sci. Ed. J., June, 99, 2001.64. Barker, P.S., Chen, J.R., Agbor, N.E., Monkman, A.P., Mars, P., and Petty, M.C.,

Sensors Actuators B, 17, 143, 1994.65. Gardner, J.W. and Bartlett, P.N., Sensors Actuators B, 18, 211, 1994.66. Casalini, R., Wilde, J.N., Nagel, J., Oertel, U., and Petty, M.C., Sensors Actuators

B, 18, 28, 1999.67. Casalini, R., Nagel, J., Oertel, U., and Petty, M.C., J. Phys. D: Appl. Phys., 31,

3146, 1998.68. Jones, C.A., Petty, M.C., and Roberts, G.G., IEEE Trans. Ultrasonics Ferroelec.

Freq. Control, 35, 736, 1988.69. Burfoot J.C. and Taylor, G.W., Polar Dielectrics and Their Applications, Macmill-

an, London, 1979.70. Richardson, T.H., in Functional Organic and Polymeric Materials, Richardson,

T.H., Ed., John Wiley & Sons, Chichester, 2000, p. 181.71. Petty, M., Tsibouklis, J., Petty, M.C., and Feast, W.J., Thin Solid Films, 210/211,

320, 1992.72. Yano, K., Ishii, T., Sano, T., Mine, T., Murai, F., Hashimoto, T., Kobayashi, T.,

Kure, T., and Seki, K., Proc. IEEE, 87, 633, 1999.73. Kouwenhoven, L.P. and McEuen, P.L., in Nanotechnology, Timp, G., Ed.,

Springer-Verlag, New York, 1999, p. 471.74. Moore, R.G.C., Evans, S.D., Shen, T., and Hodson, C.E.C., Physica E, 9, 253,

2001.75. Nabok, A.V., Richardson, T., McCartney, C., Cowlam, N., Davis, F., Stirling,

C.J.M., Ray, A.K., Gacem, V., and Gibaud, A., Thin Solid Films, 327–329, 510,1998.

76. Prasanth, N., Kumar, S., Narang, N., Major, S., Vitta, S., Talwar, S.S., Dubcek,P., Amenitsch, H., and Bernstorff, S., Colloids Surfaces A, 198–200, 59, 2002

77. Cousins, M.A. and Petty, M.C., Unpublished data ([email protected]).78. Fink, H.-W. and Schneberger, C., Nature, 398, 407, 1999.79. Dekker, C. and Ratner, M.A., Physics World, 14, 29, 2001.80. Rakitin, A., Aich, P., Papadopoulos, C., Kobzar, Y., Vedeneev, A.S., Lee, J.S.,

and Xu, J.M., Phys. Rev. Lett., 86, 3670, 2001.81. Phadke, R.S., Appl. Biochem Biotech., 96, 269, 2001.82. Rao, C.N.R. and Cheetham, A.K., J. Mater. Chem., 11, 2887, 2001.83. Adleman, L., Science, 266, 1021, 1994.



appendix:Summary of computer programs for analysis and design

Contents

Neuron modelingElectronic circuit construction and simulationData capture and processing softwareElectromagnetic modeling

What follows is list of computer software packages commonly used for thedesign and analysis of neuroprostheses and the processing of biologicalsignals. This list is by no means exhaustive but is instead intended as a basicresource for assembling a suite of analytical and design tools. The list pri-marily consists of the more commonly used packages of which the editorsare aware. Each software package is listed under the area where it is mostcommonly used.

Neuron modeling1. NEURON (http://www.neuron.yale.edu/): Free, downloadable soft-

ware for developing and exercising models of neurons and networksof neurons. The authors state that it is especially useful for problemswhere the cable properties of cells are important and where cellmembrane properties are complex.

2. GENESIS (http://www.genesis-sim.org/GENESIS/): The GEneralNEural SImulation System is a general purpose simulation platformthat supports simulations of complex models of single neuronsthrough simulations of larger networks of more abstract neuronal com-ponents. The software is free and downloadable from the website.



http://www.neuron.yale.edu/

http://www.genesis-sim.org/GENESIS/

Electronic circuit construction and simulation1. PSpice (Cadence Design Systems, Inc.): One of the most basic circuit

simulation products on the market and a common tool for electricalengineers.

2. Electronics Workbench (Electronics Workbench): Another of the mostcommonly used electronic design and automation products.(Note that many of the neuron models discussed in Chapter 3 havebeen modeled using programs such PSpice and Electronics Work-bench.)

3. Orcad Capture (Cadence Design Systems, Inc): One of a family oftools for the design of complex analog and digital circuitry for VLSIsystems.

4. Micro Magic Tools (Juniper Networks): A suite of design tools for thedesign of complex chips at the physical level. Often free for educa-tional applications.

5. Blast Software (Magma Design Automation): An innovative suite ofproducts that uses gain-based synthesis to synthesize millions ofgates simultaneously. A well-received product among VLSI re-searchers.

Data capture and processing software1. LabVIEW (National Instruments): One of the most commonly cited

data capture programs. Part of a suite of programs that performsvideo and data capture and represents capture and processing func-tions as programmable function blocks.

2. MATLAB and Simulink (The MathWorks): Multifunctional programsfor performing calculations, analyzing and visualizing data, andmodeling and simulating complex dynamic systems; contains manyspecialized toolboxes for implementing common signal processingand communications system algorithms, including wavelets.

3. IDL (Research Systems, Inc.): Another widely used program for vi-sualizing, processing and compiling data

4. Mathematica (Wolfram Research): An excellent data analysis and tech-nical programming tool, with the benefit of direct document creationfor technical presentations. Also available is the Wavelet Explorerpackage, which allows for a variety of wavelet-based processingfunctions

5. RC Electronics: A multichannel data collection program capable ofmultichannel oscilloscope and strip-chart recording and signalprocessing.

6. MSLib (www.igpm.rwth-aachen.de/urban/html/progs.html): Freewavelet processing software that can be downloaded with permis-sion and a password from the author.



http://www.igpm.rwth-aachen.de/www/index_e.html

7. Wavelet software, general (www.wavelet.org/links.html#software):A website managed by Wavelet Digest with links and information onseveral software packages for wavelet analysis.

8. THINKS (Sigma Research): Highly recommended neural networkdevelopment program with beginner and professional versions; con-sidered to be very user-friendly.

Electromagnetic modeling1. FEMLAB Electromagnetics (COMSOL AB): A component of the FEM-

LAB software for modeling and solving scientific and engineeringproblems. The electromagnetics module is capable of solving anddisplaying the results of several different electromagnetics problems.

2. FEKO (EMSS): A full-wave, method-of-moments-based program ca-pable of analyzing the electromagnetics of dielectric bodies, planarstratified media, and similar problems commonly found in neuro-prosthetic design.

3. IE3D (Zeland Software, Inc.): Also a method-of-moments-based elec-tromagnetics design and analysis tool. Zeland also offers other prod-ucts for specialized electromagnetics solutions.

4. CAEME (NSF/IEEE Center for Computer Applications in Electro-magnetics Education): A popular electromagnetics design and anal-ysis tool among academic institutions available from the NationalScience Foundation in the United States.



http://www.wavelet.org/index.php?gallery=1#software

Documents

Handbook of Neuroprosthetic Methods - W. Finn, P. LoPresti (CRC, 2003) WW